\documentclass[a4paper]{article}
\usepackage{natbib}
\usepackage{authblk}
\usepackage[latin1]{inputenc} 
\usepackage[english]{babel}
\usepackage{amsmath}
\usepackage{caption}
\usepackage{multirow}
\usepackage{xcolor}
\usepackage{url}
\usepackage[pdftex, plainpages = false, pdfpagelabels,
                pdfpagelayout = useoutlines,
                 bookmarks,
                 bookmarksopen = true,
                 bookmarksnumbered = true,
                 breaklinks = true,
                 linktocpage,
                 pagebackref,
                 colorlinks = true,
                 linkcolor = blue,
                 urlcolor  = blue,
                 citecolor = blue,
                 anchorcolor = green,
                 hyperindex = true,
                 hyperfigures]{hyperref}

\title{Creating an in-home display: experimental evidence and guidelines for design}
\author{Tamar Krishnamurti\thanks{Corresponding Author. Phone: 1-412-445-2663. Fax: Email: tamar@cmu.edu.  All materials and data, including completely reproducible statistical analyses in Sweave, can be obtained at \href{http://hdl.handle.net/1902.1/19153}{http://hdl.handle.net/1902.1/19153}. Open lab notebook can be obtained at \href{http://openwetware.org/wiki/IHD_Simulation:Notebook/IHD_Field_Trial}{http://openwetware.org/wiki/IHD_Simulation:Notebook/IHD_Field_Trial}; Applied Energy; Energy; Energy Policy; Journal of Environmental Psychology; Energy Conservation and Management; Energy and Buildings; We thank Jay apt...This material is based upon work supported by the Department of Energy under Award Numbers DE-OE0000300 and DE-OE0000204.  Disclaimer: This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights, Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or any agency thereof. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof.}}

\author{Alexander L. Davis} 
\author{Gabrielle Wong-Parodi}   
\author{Jack Wang} 
\author{Casey Canfield} 

\affil{Department of Engineering and Public Policy\\
Carnegie Mellon University}

\begin{document}
\maketitle

\begin{abstract}
In-home electricity displays (IHDs) are digital devices that can give near-real-time information about electricity usage in the home.  These devices have the potential to provide the kind of personalized feedback necessary to effect behavioral change among residential consumers.  We present an approach to in-home display design that uses research on customer preferences to determine which features to experimentally test, while comparing preferences against experimental data to determine whether people have insight into what works.  Using a computer-based simulated IHD, we found that the types of feedback information that consumers prefer (appliance-specific and dollar-feedback) are not as effective for learning than kWh feedback.  Moreover, it appears that a simpler more generalized format of information provision has the potential to be more effective than a personalized IHD for knowledge acquisition.  We discuss the need to use consumer preferences to inform IHD design but not at the cost of systematic experimental testing. (word count: xxx)}
  \\
Keywords: In-home displays; Electricity Feedback; Consumer preferences; Learning
\end{abstract}

<<preamble,echo=false,results=hide,fig=false>>=
options(scipen=0,digits=2)
@ 

To be effective, any information supplied to residential electricity customers must be adapted to their needs, especially for those who have limited knowledge of electricity.  Unfortunately, what is provided to U.S. households by electric utility companies does not meet this criterion.  Most people receive their monthly bill, scan it to identify the amount they owe, and then discard it.  This is partly because the information on the bill is in a format that is too dense or complicated to decipher \cite{kempton1994consumer}.  For example, usage information is typically presented only in kilowatt-hours (kWh), a unit that has been found to be opaque to customers \cite{darby2010smart}.  When the bill is supplemented with inserts or other offerings, they are perceived to be a scam or gouging opportunity, and are usually ignored \cite{roberts2004consumer}.  Without the right information that promotes understanding, customers cannot effectively conserve energy.

In the absence of available and usable information, customers will create `folk theories' or mental models of how electrical devices function and use energy \cite{rozenblit2002misunderstood,keil2003folkscience}.  Incorrect theories may encourage them to engage in actions that can be particularly wasteful of energy \cite{wood2003dynamic,krishnamurti2011preparing}.  Take, for example, the common folk theory about thermostats, the ``valve'' theory, which holds that the quantity of cooling or heating in the home is proportional to the thermostat setting.  This would lead customers who want to cool their household to set their thermostat lower (e.g., to zero Celsius) hoping for faster cooling, but instead just waste energy when the air conditioner cools too much.  Without easily interpretable information that corrects these folk theories, many customers would not adopt appropriate measures to save electricity, even if they wanted to.

Recognizing the problem, researchers and utility companies have tried providing customers with in-home electricity displays (IHDs) that can give near-real-time information about electricity usage.  One of the earliest examples of a simple, and particularly effective, IHD was that used in the Twin Rivers study \cite{seligman1978behavioral,seligman1981encouraging}.  In this study, participants were given a simple light that flashed blue when one could cool the home by opening the windows rather than using the air conditioner.  This display yielded almost a 20\% reduction in monthly electricity use over the short duration of the study.  Paired with the advanced meters (`smart-meters') of the smart grid, more sophisticated IHDs can provide customers high resolution feedback about their electricity consumption.

Since the seminal Twin Rivers study, mounting evidence has shown that IHDs can substantially help customers curtail their electricity use. A variety of displays have been used in these field studies, including retail (e.g., the PowerCost Monitor) and custom devices (e.g., The Residential Energy Cost Speedometer; \cite{dobson1992conservation}). Each IHD provides different types of feedback information (e.g., kWh use, cost of electricity, monthly spending), in different formats (e.g., graphs, tables, numbers, visual-analogs). In a recent review of these field trials, Davis et al. (2012) found four custom displays (Bluelight, \cite{seligman1978behavioral}; RECS \cite{dobson1992conservation}, Fitch \cite{mcclelland1979energy} and Electricity Consumption Display \cite{yun2009investigating}) to be the most effective for reducing overall consumption ($\sim$20\%, $\sim$13\%, $\sim$12\%, $\sim$12\%, respectively).  Thus, it appears that custom designed IHDs can provide the right information in an easily understood manner, leading to effective reductions in electricity use.

While these findings are encouraging, the small sample sizes of these studies ($N=$ 20, 99, 101, and 8, respectively) calls into question the real-world efficacy of these displays.  Casting further doubt, studies of popular retail devices, that have much larger sample sizes, such as the TED and PowerCost Monitor, show much smaller effect sizes (e.g., \cite{norton2008powercost}).  Moreover, once adjusted for plausible bias in study design, effects are substantially reduced \cite{davis2012setting}.  While studies that demonstrated large versus small effects differed in many ways, an important difference was the type of IHD they used, leaving open the possibility that some displays were more effective than others by design.  In addition, the existing IHD field studies vary so much in experimental methods, reporting, and features of the IHDs, it is impossible to pinpoint what made one study more successful than another.

To determine the most effective display, one would ideally test each display in the field with a large enough sample size to detect even small effects.  However, this approach is both time-consuming and prohibitively expensive, forcing utilities to guess which display would suit their customers' needs. In this paper, we present findings from a simple simulated computer-based in-home display, which allows us to test display features individually and measure their effectiveness as educational tools.

Past research on customer preferences for IHD features has used interviews, surveys, and other similar approaches (e.g., focus groups).  The options participants generated or could choose from have generally fallen into five categories \cite{wood2007energy,karjalainen2011consumer}: \emph{Units, Time Aggregation, Physical Aggregation, Comparators} and \emph{Format.}\footnote{A more exhaustive review of the specific attributes that fall into these categories may be found in Appendix~\ref{app:pref}.}

Overall, the available literature shows some agreement on what customers want in terms of units and comparators, while there are fewer findings on time aggregation, physical aggregation, and format.  However, relying on preferences to infer how customers will behave is risky, as ``what people think they want and what they actually want are not always the same'' \cite{anderson2009exploring}.  This is consistent with more basic psychological research, showing that customers are not always good at predicting what they will like, concentrating too much on changes \cite{wilson2003affective} or showing bias toward their present feelings \cite{loewenstein2003role}.  People also have been known to reject policies in prospect, but like them once implemented \cite{steg2009encouraging}.  Thus, examining preferences alone may give a certain, but potentially incorrect, perspective of how an in-home display should be designed to be most effective.

We suggest a complementary approach that uses research on customer preferences to determine what features to experimentally test, while comparing preferences against experimental data to determine whether people have insight into what works. To date, little experimental work (with the exception of enhanced bills, \cite{egan1999graphical,wilhite1999hoivik}) has been conducted.  The various field studies, interviews, and surveys have neither separated specific elements of IHDs according to their effectiveness, nor measured important intermediaries of effectiveness such as learning and motivation \cite{roberts2004consumer,wood2007energy,fischer2008feedback,roberts2003towards,neenan2009residential}. 

In the following section, we present new data on customer preferences for the attributes found on the most common commercially available IHDs. We then present findings of an experimental test of two units of information (\$ vs. kWh information) and two forms of aggregation (total vs. appliance-specific), using an IHD simulation.  We conclude with a discussion of the implications of our findings, as well as how IHD simulations can usefully inform the development and testing of IHDs in the field.

\section{Survey}
\subsection{Materials}

First, after a thorough review of all commercially available in-home displays, we developed a list of 19 displays available for direct purchase by residential homeowners.  Next, we developed a list of the most common types of electricity feedback information provided by these 19 displays and in the existing literature on IHDs.

\subsection{Procedure}

Participants were first asked to rate this set of feedback types with the following instructions: ``Here is a list of information that might appear on an in-home display. Please rate each type of information in terms of how much you would like to have it on the display.'' All information attributes were rated from 1 (not at all) to 5 (extremely), with the additional option of responding ``I don't know.''

Participants then created their own display by selecting what features they wanted from those listed above. They also responded to the following questions about the display they created, ``How much would you like to have the in-home display you created on the previous page? (1 = not at all; 5 = extremely),'' ``How effective do you think that in-home display would help you to reduce your electricity use? (1 = very ineffective, 7 = very effective),'' ``How often would you look at the in-home display you created? (1 = never, 7 = daily),'' and ``How much do you think you would save, in dollars, on your monthly electricity bill if you had the display you created on the previous page?''  The full materials are described in Appendix~\ref{app:surveymeasures}. 

\subsection{Participants}

Participants were bill-paying electricity customers ($N=151$) recruited using the Amazon Mturk system.  The average age was 31.5 years old ($SD=10.9$), with 42\% being male, and most having an income between \$51K and \$75K per year.  The average electricity bill among these customers was \$107/month.

\subsection{Results}
In general, people prefer an IHD (mean rank $=1.61$) to receiving information on a computer (mean rank $=1.93$) or smart-phone (mean rank $=2.10$).  The average willingness-to-pay for an IHD was \$150.

\subsubsection{Feedback Information Preferences}

Table~\ref{tab:info} presents the feedback information types in order of preference.  A Wilcoxon signed-rank test was used to assess whether participants ranked one attribute higher than another.  Information presented as ``bill-to-date'' and ``appliance-specific'' feedback were considered the most desirable IHD features, although neither was preferred over the other according to the Wilcoxon signed ranks test ($Z=$ \Sexpr{prettyNum(qnorm(1-(0.11)/2))}, $p=0.11$).  The least preferred way of presenting information was comparison to a ``similar household.''  This attribute was rated much less desirable than any of the other attributes ($p<0.001$).   

\begin{table}[h]
  \caption{Feedback Information Preferences}
  \label{tab:info}
  \centering
  \begin{tabular}{l c c}
   Feedback Information Type & Mean & SD \\ \hline
   Bill-to-date [IA1] & 4.11 & 0.94 \\
   Appliance-Specific [IA2] & 4.00 & 0.96 \\ 
   Daily Projections [IA3] & 3.83 & 0.99 \\
   Monthly Projections [IA4] & 3.80 & 0.96 \\
   kWh-to-date  [IA5] & 3.80 & 0.96 \\
   Daily price [IA6] & 3.70 & 1.16 \\
   Daily peak use [IA7] & 3.69 & 1.04 \\
   Monthly peak use [IA8] & 3.68 & 1.08 \\ 
   Goal tracking [IA9] & 3.61 & 0.96 \\
   Greenness [IA10] & 3.40 & 1.10 \\
   Similar Household [IA11] & 2.87 & 1.20 \\ \hline
   \end{tabular}
  \end{table}
   
\subsubsection{Create-Your Own}

Participants' constructions of their own IHD were almost identical to their ratings of individual feedback information types, so we omit them here.  On average participants expected to like the display they'd created strongly, as seen in the mean rating being significantly above the scale midpoint of 3.0 in a one-sample t-test, $t(x)=$ x, $p<$ .001.  They thought it would be effective way for them to reduce their electricity use with a mean judged effectiveness significantly above the scale midpoint of 4.0, $t(x)=$ x, $p<$ .001, and expected to look at it on average 2–3 times a week ($M=$ 6.11, $SD=$ 1.1, $N=$ 138). They anticipated an average savings on their monthly bill of \$25 ($SD=$ 29, $N=$ 131), with the average monthly electricity bill being \$106 ($SD=$ 81, $N=$ 106).

\subsection{Discussion}

Bill-to-date and appliance-specific feedback stood out as the two most preferred types of information.  This finding is consistent with previous research on preferences for IHD content \cite{darby2010smart,karjalainen2011consumer,anderson2009exploring}.  There was little differentiation between other types of information, such as projections or goals.  However, the more `gimmicky' features, greenness and social comparisons, were strongly disliked.  Participants had unreasonably high expectations for monthly savings from the IHD (about 25\% of their monthly bill), given that the average monthly savings from IHDs has never been above 20\% in previous research (Davis et al, 2012). 

Applied energy cites XX.

Our survey suggests that people want to be able to simply look at an IHD and obtain easy-to-interpret feedback information to help them manage their monthly budget and control their current consumption. Our next study draws on these preferences to test how well people learn from what they say they want. Specifically, our study and previous research strongly suggests that people prefer bill-to-date information and appliance-specific feedback in units of \$. However, do they actually benefit from this configuration, or is it possible to get the same benefit using the simpler to implement kWh and bill-to-date configuration? The experiments described in the next section try to answer this question using a simulated in-home display.

\section{Simulated In-Home Display Experiment}
\subsection{Methods}
Participants interacted with a computer-based in-home display simulation that allowed them to turn the appliances on and off, change the settings on various thermostats, and alter the length of use for each in 30 minute incrememnts in the house of a fictional family, ``The Smiths.''  Feedback information on electricity use was updated according to these manipulations and presented in a tabular format, as seen in Figure~\ref{fig:simpic}.

We selected the 11 most commonly owned electricity consuming home appliances in the U.S. for participants to manipulate the use of (ref): (1) air conditioner, (2) water heater, (3) indoor lights, (4) outdoor lights, (5) refrigerator, (6) freezer, (7) oven, (8) microwave, (9) television, (10) washing machine, and (11) dryer.  We used standard estimates for the real-time electricity consumption values, as well as used the standard current estimate for the average cost of electricity in the US (ref). These values for the 11 appliances were used as inputs for our simulated IHD, and thus were reflected in the values seen by our participants.  

\begin{figure}[h]
    \centering
\scalebox{1.2}{\includegraphics{simpic}}
\caption{Screenshot of the simulated in-home display.}
\label{fig:simpic}
\end{figure}

\subsection{Experimental Conditions}

Participants were provided the same electricity knowledge test, before and after interacting with the simulated IHD. There were six in-home display simulations that each provided participants with different information when they manipulated the appliances. The specific conditions were designed to test the effectiveness on knowledge acquisition of the information that people reported wanting to see (appliance-specific feedback in \$ units) vs. the information that would be easiest for a utility to provide (aggregate kwH usage). Details of each condition can be seen in Table~\ref{tab:conditions}. Knowledge acquisition was assessed on ten out of eleven appliances.\footnote{Although participants were able to manipulate and received feedback on the electricity consumption of  11 appliances, pre-testing showed that participants felt much more comfortable with 10-item rankings. The difference in usage between indoor and outdoor lights were marginal compared to the other appliances, therefore outdoor lights were removed from the ranking tasks.}

\begin{table}[h]
  \caption{Description of feedback information for the 6 conditions and 1 control group.\footnote{For all conditions providing cost information, a kWh was priced at \$0.13.}}
  \label{tab:conditions}
  \centering
  \begin{tabular}{c p{7cm}}
    Group Condition & Feedback Information \\ \hline
    kWh only & Participants were shown kWh feedback information used for running the appliances at the settings they made for the duration they specified. \\
    \$ only & Participants were shown cost (\$) feedback information used for running the appliances at the settings they made for the duration they specified.\\
\$ and kWh & Participants were shown both kWh and cost (\$) feedback information used for running the appliances at the settings they made for the duration they specified. \\
\$ by appliance & Participants were shown kWh feedback information in total, as well as for each appliance, used for running the appliances at the settings they made for the duration they specified. \\
kWh by appliance & Participants were shown cost (\$) feedback information in total, as well as for each appliance, used for running the appliances at the settings they made for the duration they specified. \\
\$ and kWh by appliance & Participants were shown kWh and cost (\$) feedback information in total, as well as for each appliance, used for running the appliances at the settings they made for the duration they specified. \\
Passive Learning & Participants were given the answers to the electricity knowledge pre-test and then asked to provide those answers in the post-test. \\ \hline
  \end{tabular}
  \end{table}

We also included a control condition simulating more generic educational materials in which, customers are simply told to use less electricity, or are provided with information on how much electricity appliances use in general \cite{wood2003dynamic}.  In our \emph{passive learning} control condition, participants were given the answers to the electricity knowledge pre-test and then asked to provide those answers in the post-test.  This type of passive, as opposed to discovery, learning can come from a variety of sources such as ``educational campaigns, advertisements, advisory services and news media'' \cite{darby2010smart}.

Both theory and practice suggest that that passive learning is not sufficient for behavior change, although it may be more effective in cases when behavior change is convenient and cheap (low barriers) \cite{steg2009encouraging}.  However, here passive learning serves as a control by establishing whether participants can process the knowledge areas that are being tested given that aspects of this topic may be something that participants have had little or no prior exposure to.

<<exp1,echo=false,results=hide,fig=false>>=
c<-c(1,22:51,70:89,90:94,245:264,265:269,282:298,333)

exp1.t<-read.csv("simdata1.csv",na.string="",skip=1,header=TRUE,stringsAsFactors=FALSE)
exp1<-exp1.t[,c]
g1<-c("ResponseID","AC.use.reg","water.heater.use.reg","fridge.use.reg","freezer.use.reg","microwave.use.reg","washer.use.reg","oven.use.reg","stove.use.reg","TV.use.reg","dryer.use.reg","AC.pwr.src","water.heater.pwr.src","fridge.pwr.src","freezer.pwr.src","microwave.pwr.src","washer.pwr.src","oven.pwr.src","stove.pwr.src","TV.pwr.src","dryer.pwr.src","AC.num","water.heater.num","fridge.num","freezer.num","microwave.num","washer.num","oven.num","stove.num","TV.num","dryer.num","freezer.rank.use.pre","fridge.rank.use.pre","indoor.rank.use.pre","oven.rank.use.pre","TV.rank.use.pre","microwave.rank.use.pre","dryer.rank.use.pre","AC.rank.use.pre","water.heater.rank.use.pre","washer.rank.use.pre","freezer.rank.cost.pre","fridge.rank.cost.pre","indoor.rank.cost.pre","oven.rank.cost.pre","TV.rank.cost.pre","microwave.rank.cost.pre","dryer.rank.cost.pre","AC.rank.cost.pre","water.heater.rank.cost.pre","washer.rank.cost.pre","electricity.units.pre","energy.calc.pre","kwh.cost.pre.doll","kwh.cost.pre.cents","kwh.use.pre","freezer.rank.use.post","fridge.rank.use.post","indoor.rank.use.post","oven.rank.use.post","TV.rank.use.post","microwave.rank.use.post","dryer.rank.use.post","AC.rank.use.post","water.heater.rank.use.post","washer.rank.use.post","freezer.rank.cost.post","fridge.rank.cost.post","indoor.rank.cost.post","oven.rank.cost.post","TV.rank.cost.post","microwave.rank.cost.post","dryer.rank.cost.post","AC.rank.cost.post","water.heater.rank.cost.post","washer.rank.cost.post","electricity.units.post","energy.calc.post","kwh.cost.post.doll","kwh.cost.post.cents","kwh.use.post","feature.add.remove.open","total.kwh","total.cost","kwh.appl","cost.appl","helpfulness","helpfulness.open","reduction.kwh","reduction.kwh.open","learning","learning.open","wtp","wtp.open","expected.monthly.savings","expected.monthly.savings.open","gender","age","condition")

names(exp1)<-g1

##Use rankings by appliance##
exp1$AC.true.use<-1
exp1$dryer.true.use<-2
exp1$oven.true.use<-3
exp1$micro.true.use<-4
exp1$heater.true.use<-5
exp1$washer.true.use<-6
exp1$freezer.true.use<-7
exp1$TV.true.use<-8
exp1$fridge.true.use<-9
exp1$indoor.true.use<-10

exp1$AC.pre.dev.use<-abs(1-exp1$AC.rank.use.pre)
exp1$dryer.pre.dev.use<-abs(2-exp1$dryer.rank.use.pre)
exp1$oven.pre.dev.use<-abs(3-exp1$oven.rank.use.pre)
exp1$micro.pre.dev.use<-abs(4-exp1$microwave.rank.use.pre)
exp1$heater.pre.dev.use<-abs(5-exp1$water.heater.rank.use.pre)
exp1$washer.pre.dev.use<-abs(6-exp1$washer.rank.use.pre)
exp1$freezer.pre.dev.use<-abs(7-exp1$freezer.rank.use.pre)
exp1$TV.pre.dev.use<-abs(8-exp1$TV.rank.use.pre)
exp1$fridge.pre.dev.use<-abs(9-exp1$fridge.rank.use.pre)
exp1$indoor.pre.dev.use<-abs(10-exp1$indoor.rank.use.pre)                      

exp1$AC.post.dev.use<-abs(1-exp1$AC.rank.use.post)
exp1$dryer.post.dev.use<-abs(2-exp1$dryer.rank.use.post)
exp1$oven.post.dev.use<-abs(3-exp1$oven.rank.use.post)
exp1$micro.post.dev.use<-abs(4-exp1$microwave.rank.use.post)
exp1$heater.post.dev.use<-abs(5-exp1$water.heater.rank.use.post)
exp1$washer.post.dev.use<-abs(6-exp1$washer.rank.use.post)
exp1$freezer.post.dev.use<-abs(7-exp1$freezer.rank.use.post)
exp1$TV.post.dev.use<-abs(8-exp1$TV.rank.use.post)
exp1$fridge.post.dev.use<-abs(9-exp1$fridge.rank.use.post)
exp1$indoor.post.dev.use<-abs(10-exp1$indoor.rank.use.post)                     

pre.diff.use.AC<-lm(AC.rank.use.pre~condition,data=exp1)
pre.diff.use.dryer<-lm(dryer.rank.use.pre~condition,data=exp1)
pre.diff.use.oven<-lm(oven.rank.use.pre~condition,data=exp1)
pre.diff.use.micro<-lm(microwave.rank.use.pre~condition,data=exp1)
pre.diff.use.heater<-lm(water.heater.rank.use.pre~condition,data=exp1)
pre.diff.use.washer<-lm(washer.rank.use.pre~condition,data=exp1)
pre.diff.use.freezer<-lm(freezer.rank.use.pre~condition,data=exp1)
pre.diff.use.TV<-lm(TV.rank.use.pre~condition,data=exp1)
pre.diff.use.fridge<-lm(fridge.rank.use.pre~condition,data=exp1)
pre.diff.use.indoor<-lm(indoor.rank.use.pre~condition,data=exp1)

##AC pre-post signed ranks tests for use##

rdACu<-wilcox.test(exp1$AC.pre.dev.use,exp1$AC.post.dev.use,paired=TRUE,data=exp1)
ACU1<-wilcox.test(exp1$AC.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$AC.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
ACU2<-wilcox.test(exp1$AC.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$AC.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
ACU3<-wilcox.test(exp1$AC.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$AC.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
ACU4<-wilcox.test(exp1$AC.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$AC.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
ACU5<-wilcox.test(exp1$AC.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$AC.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
ACU6<-wilcox.test(exp1$AC.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$AC.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
ACU7<-wilcox.test(exp1$AC.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$AC.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

AC1.pr<-exp1$AC.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
AC2.pr<-exp1$AC.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
AC3.pr<-exp1$AC.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
AC4.pr<-exp1$AC.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
AC5.pr<-exp1$AC.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
AC6.pr<-exp1$AC.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
AC7.pr<-exp1$AC.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]
AC1.po<-exp1$AC.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
AC2.po<-exp1$AC.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
AC3.po<-exp1$AC.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
AC4.po<-exp1$AC.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
AC5.po<-exp1$AC.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
AC6.po<-exp1$AC.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
AC7.po<-exp1$AC.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]

#AC5.xx<-na.omit(exp1[exp1$condition==levels(as.factor(exp1$condition))[5],c("ResponseID","AC.pre.dev.use")])
#AC5.xy<-AC5.xx[-36,]
##r6hn missing
#aa<-mean(AC5.xy[,2])
#AC5.xz<-na.omit(exp1[exp1$condition==levels(as.factor(exp1$condition))[5],c("ResponseID","AC.post.dev.use")])
#AC5.zy<-AC5.xz[-29,]
#aa<-mean(AC5.zy[,2])

#AC4.xx<-na.omit(exp1[exp1$condition==levels(as.factor(exp1$condition))[4],c("ResponseID","AC.pre.dev.use")])
#AC4.xy<-AC4.xx[-36,]
#aa<-mean(AC5.xy[,2])
#AC4.xz<-na.omit(exp1[exp1$condition==levels(as.factor(exp1$condition))[4],c("ResponseID","AC.post.dev.use")])
#AC4.zy<-AC4.xz[-29,]
#aa<-mean(AC4.zy[,2])

mAC1.pr<-mean(AC1.pr,na.rm=TRUE)
mAC1.po<-mean(AC1.po,na.rm=TRUE)
mAC2.pr<-mean(AC2.pr,na.rm=TRUE)
mAC2.po<-mean(AC2.po,na.rm=TRUE)
mAC3.pr<-mean(AC3.pr,na.rm=TRUE)
mAC3.po<-mean(AC3.po,na.rm=TRUE)
mAC4.pr<-mean(AC4.pr,na.rm=TRUE)
mAC4.po<-mean(AC4.po,na.rm=TRUE)
mAC5.pr<-mean(AC5.pr,na.rm=TRUE)
mAC5.po<-mean(AC5.po,na.rm=TRUE)
mAC6.pr<-mean(AC6.pr,na.rm=TRUE)
mAC6.po<-mean(AC6.po,na.rm=TRUE)
mAC7.pr<-mean(AC7.pr,na.rm=TRUE)
mAC7.po<-mean(AC7.po,na.rm=TRUE)

##Dryer pre-post signed ranks tests for use##

rddryeru<-wilcox.test(exp1$dryer.pre.dev.use,exp1$dryer.post.dev.use,paired=TRUE,data=exp1)
DRU1<-wilcox.test(exp1$dryer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$dryer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
DRU2<-wilcox.test(exp1$dryer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$dryer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
DRU3<-wilcox.test(exp1$dryer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$dryer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
DRU4<-wilcox.test(exp1$dryer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$dryer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
DRU5<-wilcox.test(exp1$dryer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$dryer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
DRU6<-wilcox.test(exp1$dryer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$dryer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
DRU7<-wilcox.test(exp1$dryer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$dryer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

dry1.pr<-exp1$dryer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
dry2.pr<-exp1$dryer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
dry3.pr<-exp1$dryer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
dry4.pr<-exp1$dryer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
dry5.pr<-exp1$dryer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
dry6.pr<-exp1$dryer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
dry7.pr<-exp1$dryer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]
dry1.po<-exp1$dryer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
dry2.po<-exp1$dryer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
dry3.po<-exp1$dryer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
dry4.po<-exp1$dryer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
dry5.po<-exp1$dryer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
dry6.po<-exp1$dryer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
dry7.po<-exp1$dryer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]

mdry1.pr<-mean(dry1.pr,na.rm=TRUE)
mdry1.po<-mean(dry1.po,na.rm=TRUE)
mdry2.pr<-mean(dry2.pr,na.rm=TRUE)
mdry2.po<-mean(dry2.po,na.rm=TRUE)
mdry3.pr<-mean(dry3.pr,na.rm=TRUE)
mdry3.po<-mean(dry3.po,na.rm=TRUE)
mdry4.pr<-mean(dry4.pr,na.rm=TRUE)
mdry4.po<-mean(dry4.po,na.rm=TRUE)
mdry5.pr<-mean(dry5.pr,na.rm=TRUE)
mdry5.po<-mean(dry5.po,na.rm=TRUE)
mdry6.pr<-mean(dry6.pr,na.rm=TRUE)
mdry6.po<-mean(dry6.po,na.rm=TRUE)
mdry7.pr<-mean(dry7.pr,na.rm=TRUE)
mdry7.po<-mean(dry7.po,na.rm=TRUE)

##Oven pre-post signed ranks tests for use##

rdovenu<-wilcox.test(exp1$oven.pre.dev.use,exp1$oven.post.dev.use,paired=TRUE,data=exp1)
OVU1<-wilcox.test(exp1$oven.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$oven.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
OVU2<-wilcox.test(exp1$oven.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$oven.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
OVU3<-wilcox.test(exp1$oven.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$oven.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
OVU4<-wilcox.test(exp1$oven.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$oven.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
OVU5<-wilcox.test(exp1$oven.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$oven.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
OVU6<-wilcox.test(exp1$oven.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$oven.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
OVU7<-wilcox.test(exp1$oven.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$oven.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

ov1.pr<-exp1$oven.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
ov2.pr<-exp1$oven.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
ov3.pr<-exp1$oven.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
ov4.pr<-exp1$oven.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
ov5.pr<-exp1$oven.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
ov6.pr<-exp1$oven.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
ov7.pr<-exp1$oven.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]
ov1.po<-exp1$oven.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
ov2.po<-exp1$oven.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
ov3.po<-exp1$oven.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
ov4.po<-exp1$oven.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
ov5.po<-exp1$oven.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
ov6.po<-exp1$oven.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
ov7.po<-exp1$oven.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]

mov1.pr<-mean(ov1.pr,na.rm=TRUE)
mov1.po<-mean(ov1.po,na.rm=TRUE)
mov2.pr<-mean(ov2.pr,na.rm=TRUE)
mov2.po<-mean(ov2.po,na.rm=TRUE)
mov3.pr<-mean(ov3.pr,na.rm=TRUE)
mov3.po<-mean(ov3.po,na.rm=TRUE)
mov4.pr<-mean(ov4.pr,na.rm=TRUE)
mov4.po<-mean(ov4.po,na.rm=TRUE)
mov5.pr<-mean(ov5.pr,na.rm=TRUE)
mov5.po<-mean(ov5.po,na.rm=TRUE)
mov6.pr<-mean(ov6.pr,na.rm=TRUE)
mov6.po<-mean(ov6.po,na.rm=TRUE)
mov7.pr<-mean(ov7.pr,na.rm=TRUE)
mov7.po<-mean(ov7.po,na.rm=TRUE)

##Microwave pre-post signed ranks tests for use##

rdmicrou<-wilcox.test(exp1$micro.pre.dev.use,exp1$micro.post.dev.use,paired=TRUE,data=exp1)
MIU1<-wilcox.test(exp1$micro.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$micro.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
MIU2<-wilcox.test(exp1$micro.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$micro.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
MIU3<-wilcox.test(exp1$micro.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$micro.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
MIU4<-wilcox.test(exp1$micro.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$micro.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
MIU5<-wilcox.test(exp1$micro.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$micro.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
MIU6<-wilcox.test(exp1$micro.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$micro.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
MIU7<-wilcox.test(exp1$micro.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$micro.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

mi1.pr<-exp1$micro.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
mi2.pr<-exp1$micro.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
mi3.pr<-exp1$micro.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
mi4.pr<-exp1$micro.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
mi5.pr<-exp1$micro.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
mi6.pr<-exp1$micro.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
mi7.pr<-exp1$micro.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]
mi1.po<-exp1$micro.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
mi2.po<-exp1$micro.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
mi3.po<-exp1$micro.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
mi4.po<-exp1$micro.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
mi5.po<-exp1$micro.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
mi6.po<-exp1$micro.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
mi7.po<-exp1$micro.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]

mmi1.pr<-mean(mi1.pr,na.rm=TRUE)
mmi1.po<-mean(mi1.po,na.rm=TRUE)
mmi2.pr<-mean(mi2.pr,na.rm=TRUE)
mmi2.po<-mean(mi2.po,na.rm=TRUE)
mmi3.pr<-mean(mi3.pr,na.rm=TRUE)
mmi3.po<-mean(mi3.po,na.rm=TRUE)
mmi4.pr<-mean(mi4.pr,na.rm=TRUE)
mmi4.po<-mean(mi4.po,na.rm=TRUE)
mmi5.pr<-mean(mi5.pr,na.rm=TRUE)
mmi5.po<-mean(mi5.po,na.rm=TRUE)
mmi6.pr<-mean(mi6.pr,na.rm=TRUE)
mmi6.po<-mean(mi6.po,na.rm=TRUE)
mmi7.pr<-mean(mi7.pr,na.rm=TRUE)
mmi7.po<-mean(mi7.po,na.rm=TRUE)

##Heater pre-post signed ranks tests for use##

rdheateru<-wilcox.test(exp1$heater.pre.dev.use,exp1$heater.post.dev.use,paired=TRUE,data=exp1)
HEU1<-wilcox.test(exp1$heater.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$heater.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
HEU2<-wilcox.test(exp1$heater.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$heater.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
HEU3<-wilcox.test(exp1$heater.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$heater.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
HEU4<-wilcox.test(exp1$heater.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$heater.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
HEU5<-wilcox.test(exp1$heater.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$heater.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
HEU6<-wilcox.test(exp1$heater.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$heater.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
HEU7<-wilcox.test(exp1$heater.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$heater.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

he1.pr<-exp1$heater.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
he2.pr<-exp1$heater.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
he3.pr<-exp1$heater.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
he4.pr<-exp1$heater.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
he5.pr<-exp1$heater.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
he6.pr<-exp1$heater.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
he7.pr<-exp1$heater.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]
he1.po<-exp1$heater.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
he2.po<-exp1$heater.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
he3.po<-exp1$heater.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
he4.po<-exp1$heater.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
he5.po<-exp1$heater.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
he6.po<-exp1$heater.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
he7.po<-exp1$heater.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]

mhe1.pr<-mean(he1.pr,na.rm=TRUE)
mhe1.po<-mean(he1.po,na.rm=TRUE)
mhe2.pr<-mean(he2.pr,na.rm=TRUE)
mhe2.po<-mean(he2.po,na.rm=TRUE)
mhe3.pr<-mean(he3.pr,na.rm=TRUE)
mhe3.po<-mean(he3.po,na.rm=TRUE)
mhe4.pr<-mean(he4.pr,na.rm=TRUE)
mhe4.po<-mean(he4.po,na.rm=TRUE)
mhe5.pr<-mean(he5.pr,na.rm=TRUE)
mhe5.po<-mean(he5.po,na.rm=TRUE)
mhe6.pr<-mean(he6.pr,na.rm=TRUE)
mhe6.po<-mean(he6.po,na.rm=TRUE)
mhe7.pr<-mean(he7.pr,na.rm=TRUE)
mhe7.po<-mean(he7.po,na.rm=TRUE)

##Washer pre-post signed ranks tests for use##

rdwasheru<-wilcox.test(exp1$washer.pre.dev.use,exp1$washer.post.dev.use,paired=TRUE,data=exp1)
WAU1<-wilcox.test(exp1$washer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$washer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
WAU2<-wilcox.test(exp1$washer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$washer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
WAU3<-wilcox.test(exp1$washer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$washer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
WAU4<-wilcox.test(exp1$washer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$washer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
WAU5<-wilcox.test(exp1$washer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$washer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
WAU6<-wilcox.test(exp1$washer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$washer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
WAU7<-wilcox.test(exp1$washer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$washer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

wa1.pr<-exp1$washer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
wa2.pr<-exp1$washer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
wa3.pr<-exp1$washer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
wa4.pr<-exp1$washer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
wa5.pr<-exp1$washer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
wa6.pr<-exp1$washer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
wa7.pr<-exp1$washer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]
wa1.po<-exp1$washer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
wa2.po<-exp1$washer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
wa3.po<-exp1$washer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
wa4.po<-exp1$washer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
wa5.po<-exp1$washer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
wa6.po<-exp1$washer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
wa7.po<-exp1$washer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]

mwa1.pr<-mean(wa1.pr,na.rm=TRUE)
mwa1.po<-mean(wa1.po,na.rm=TRUE)
mwa2.pr<-mean(wa2.pr,na.rm=TRUE)
mwa2.po<-mean(wa2.po,na.rm=TRUE)
mwa3.pr<-mean(wa3.pr,na.rm=TRUE)
mwa3.po<-mean(wa3.po,na.rm=TRUE)
mwa4.pr<-mean(wa4.pr,na.rm=TRUE)
mwa4.po<-mean(wa4.po,na.rm=TRUE)
mwa5.pr<-mean(wa5.pr,na.rm=TRUE)
mwa5.po<-mean(wa5.po,na.rm=TRUE)
mwa6.pr<-mean(wa6.pr,na.rm=TRUE)
mwa6.po<-mean(wa6.po,na.rm=TRUE)
mwa7.pr<-mean(wa7.pr,na.rm=TRUE)
mwa7.po<-mean(wa7.po,na.rm=TRUE)

##Freezer pre-post signed ranks tests for use##

rdfreezeru<-wilcox.test(exp1$freezer.pre.dev.use,exp1$freezer.post.dev.use,paired=TRUE,data=exp1)
FRU1<-wilcox.test(exp1$freezer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$freezer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
FRU2<-wilcox.test(exp1$freezer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$freezer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
FRU3<-wilcox.test(exp1$freezer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$freezer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
FRU4<-wilcox.test(exp1$freezer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$freezer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
FRU5<-wilcox.test(exp1$freezer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$freezer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
FRU6<-wilcox.test(exp1$freezer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$freezer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
FRU7<-wilcox.test(exp1$freezer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$freezer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

fr1.pr<-exp1$freezer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
fr2.pr<-exp1$freezer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
fr3.pr<-exp1$freezer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
fr4.pr<-exp1$freezer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
fr5.pr<-exp1$freezer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
fr6.pr<-exp1$freezer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
fr7.pr<-exp1$freezer.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]
fr1.po<-exp1$freezer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
fr2.po<-exp1$freezer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
fr3.po<-exp1$freezer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
fr4.po<-exp1$freezer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
fr5.po<-exp1$freezer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
fr6.po<-exp1$freezer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
fr7.po<-exp1$freezer.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]

mfr1.pr<-mean(fr1.pr,na.rm=TRUE)
mfr1.po<-mean(fr1.po,na.rm=TRUE)
mfr2.pr<-mean(fr2.pr,na.rm=TRUE)
mfr2.po<-mean(fr2.po,na.rm=TRUE)
mfr3.pr<-mean(fr3.pr,na.rm=TRUE)
mfr3.po<-mean(fr3.po,na.rm=TRUE)
mfr4.pr<-mean(fr4.pr,na.rm=TRUE)
mfr4.po<-mean(fr4.po,na.rm=TRUE)
mfr5.pr<-mean(fr5.pr,na.rm=TRUE)
mfr5.po<-mean(fr5.po,na.rm=TRUE)
mfr6.pr<-mean(fr6.pr,na.rm=TRUE)
mfr6.po<-mean(fr6.po,na.rm=TRUE)
mfr7.pr<-mean(fr7.pr,na.rm=TRUE)
mfr7.po<-mean(fr7.po,na.rm=TRUE)

##TV pre-post signed ranks tests for use##

rdTVu<-wilcox.test(exp1$TV.pre.dev.use,exp1$TV.post.dev.use,paired=TRUE,data=exp1)
TVU1<-wilcox.test(exp1$TV.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$TV.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
TVU2<-wilcox.test(exp1$TV.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$TV.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
TVU3<-wilcox.test(exp1$TV.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$TV.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
TVU4<-wilcox.test(exp1$TV.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$TV.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
TVU5<-wilcox.test(exp1$TV.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$TV.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
TVU6<-wilcox.test(exp1$TV.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$TV.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
TVU7<-wilcox.test(exp1$TV.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$TV.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

TV1.pr<-exp1$TV.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
TV2.pr<-exp1$TV.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
TV3.pr<-exp1$TV.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
TV4.pr<-exp1$TV.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
TV5.pr<-exp1$TV.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
TV6.pr<-exp1$TV.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
TV7.pr<-exp1$TV.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]
TV1.po<-exp1$TV.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
TV2.po<-exp1$TV.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
TV3.po<-exp1$TV.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
TV4.po<-exp1$TV.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
TV5.po<-exp1$TV.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
TV6.po<-exp1$TV.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
TV7.po<-exp1$TV.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]

mTV1.pr<-mean(TV1.pr,na.rm=TRUE)
mTV1.po<-mean(TV1.po,na.rm=TRUE)
mTV2.pr<-mean(TV2.pr,na.rm=TRUE)
mTV2.po<-mean(TV2.po,na.rm=TRUE)
mTV3.pr<-mean(TV3.pr,na.rm=TRUE)
mTV3.po<-mean(TV3.po,na.rm=TRUE)
mTV4.pr<-mean(TV4.pr,na.rm=TRUE)
mTV4.po<-mean(TV4.po,na.rm=TRUE)
mTV5.pr<-mean(TV5.pr,na.rm=TRUE)
mTV5.po<-mean(TV5.po,na.rm=TRUE)
mTV6.pr<-mean(TV6.pr,na.rm=TRUE)
mTV6.po<-mean(TV6.po,na.rm=TRUE)
mTV7.pr<-mean(TV7.pr,na.rm=TRUE)
mTV7.po<-mean(TV7.po,na.rm=TRUE)

##Fridge pre-post signed ranks tests for use##

rdfridgeu<-wilcox.test(exp1$fridge.pre.dev.use,exp1$fridge.post.dev.use,paired=TRUE,data=exp1)
FDU1<-wilcox.test(exp1$fridge.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$fridge.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
FDU2<-wilcox.test(exp1$fridge.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$fridge.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
FDU3<-wilcox.test(exp1$fridge.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$fridge.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
FDU4<-wilcox.test(exp1$fridge.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$fridge.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
FDU5<-wilcox.test(exp1$fridge.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$fridge.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
FDU6<-wilcox.test(exp1$fridge.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$fridge.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
FDU7<-wilcox.test(exp1$fridge.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$fridge.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

FD1.pr<-exp1$fridge.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
FD2.pr<-exp1$fridge.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
FD3.pr<-exp1$fridge.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
FD4.pr<-exp1$fridge.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
FD5.pr<-exp1$fridge.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
FD6.pr<-exp1$fridge.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
FD7.pr<-exp1$fridge.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]
FD1.po<-exp1$fridge.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
FD2.po<-exp1$fridge.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
FD3.po<-exp1$fridge.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
FD4.po<-exp1$fridge.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
FD5.po<-exp1$fridge.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
FD6.po<-exp1$fridge.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
FD7.po<-exp1$fridge.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]

mFD1.pr<-mean(FD1.pr,na.rm=TRUE)
mFD1.po<-mean(FD1.po,na.rm=TRUE)
mFD2.pr<-mean(FD2.pr,na.rm=TRUE)
mFD2.po<-mean(FD2.po,na.rm=TRUE)
mFD3.pr<-mean(FD3.pr,na.rm=TRUE)
mFD3.po<-mean(FD3.po,na.rm=TRUE)
mFD4.pr<-mean(FD4.pr,na.rm=TRUE)
mFD4.po<-mean(FD4.po,na.rm=TRUE)
mFD5.pr<-mean(FD5.pr,na.rm=TRUE)
mFD5.po<-mean(FD5.po,na.rm=TRUE)
mFD6.pr<-mean(FD6.pr,na.rm=TRUE)
mFD6.po<-mean(FD6.po,na.rm=TRUE)
mFD7.pr<-mean(FD7.pr,na.rm=TRUE)
mFD7.po<-mean(FD7.po,na.rm=TRUE)

##Indoor pre-post signed ranks tests for use##

rdindooru<-wilcox.test(exp1$indoor.pre.dev.use,exp1$indoor.post.dev.use,paired=TRUE,data=exp1)
INU1<-wilcox.test(exp1$indoor.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$indoor.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
INU2<-wilcox.test(exp1$indoor.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$indoor.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
INU3<-wilcox.test(exp1$indoor.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$indoor.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
INU4<-wilcox.test(exp1$indoor.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$indoor.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
INU5<-wilcox.test(exp1$indoor.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$indoor.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
INU6<-wilcox.test(exp1$indoor.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$indoor.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
INU7<-wilcox.test(exp1$indoor.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$indoor.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

IN1.pr<-exp1$indoor.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
IN2.pr<-exp1$indoor.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
IN3.pr<-exp1$indoor.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
IN4.pr<-exp1$indoor.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
IN5.pr<-exp1$indoor.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
IN6.pr<-exp1$indoor.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
IN7.pr<-exp1$indoor.pre.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]
IN1.po<-exp1$indoor.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[1]]
IN2.po<-exp1$indoor.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[2]]
IN3.po<-exp1$indoor.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[3]]
IN4.po<-exp1$indoor.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[4]]
IN5.po<-exp1$indoor.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[5]]
IN6.po<-exp1$indoor.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[6]]
IN7.po<-exp1$indoor.post.dev.use[exp1$condition==levels(as.factor(exp1$condition))[7]]

mIN1.pr<-mean(IN1.pr,na.rm=TRUE)
mIN1.po<-mean(IN1.po,na.rm=TRUE)
mIN2.pr<-mean(IN2.pr,na.rm=TRUE)
mIN2.po<-mean(IN2.po,na.rm=TRUE)
mIN3.pr<-mean(IN3.pr,na.rm=TRUE)
mIN3.po<-mean(IN3.po,na.rm=TRUE)
mIN4.pr<-mean(IN4.pr,na.rm=TRUE)
mIN4.po<-mean(IN4.po,na.rm=TRUE)
mIN5.pr<-mean(IN5.pr,na.rm=TRUE)
mIN5.po<-mean(IN5.po,na.rm=TRUE)
mIN6.pr<-mean(IN6.pr,na.rm=TRUE)
mIN6.po<-mean(IN6.po,na.rm=TRUE)
mIN7.pr<-mean(IN7.pr,na.rm=TRUE)
mIN7.po<-mean(IN7.po,na.rm=TRUE)

####
##cost rankings by appliance##
exp1$AC.true.cost<-1
exp1$heater.true.cost<-2
exp1$freezer.true.cost<-3
exp1$fridge.true.cost<-4
exp1$dryer.true.cost<-5
exp1$oven.true.cost<-6
exp1$TV.true.cost<-7
exp1$micro.true.cost<-8
exp1$washer.true.cost<-9
exp1$indoor.true.cost<-10

exp1$AC.pre.dev.cost<-abs(1-exp1$AC.rank.cost.pre)
exp1$dryer.pre.dev.cost<-abs(5-exp1$dryer.rank.cost.pre)
exp1$oven.pre.dev.cost<-abs(6-exp1$oven.rank.cost.pre)
exp1$micro.pre.dev.cost<-abs(8-exp1$microwave.rank.cost.pre)
exp1$heater.pre.dev.cost<-abs(2-exp1$water.heater.rank.cost.pre)
exp1$washer.pre.dev.cost<-abs(9-exp1$washer.rank.cost.pre)
exp1$freezer.pre.dev.cost<-abs(3-exp1$freezer.rank.cost.pre)
exp1$TV.pre.dev.cost<-abs(7-exp1$TV.rank.cost.pre)
exp1$fridge.pre.dev.cost<-abs(4-exp1$fridge.rank.cost.pre)
exp1$indoor.pre.dev.cost<-abs(10-exp1$indoor.rank.cost.pre)                     

exp1$AC.post.dev.cost<-abs(1-exp1$AC.rank.cost.post)
exp1$dryer.post.dev.cost<-abs(5-exp1$dryer.rank.cost.post)
exp1$oven.post.dev.cost<-abs(6-exp1$oven.rank.cost.post)
exp1$micro.post.dev.cost<-abs(8-exp1$microwave.rank.cost.post)
exp1$heater.post.dev.cost<-abs(2-exp1$water.heater.rank.cost.post)
exp1$washer.post.dev.cost<-abs(9-exp1$washer.rank.cost.post)
exp1$freezer.post.dev.cost<-abs(3-exp1$freezer.rank.cost.post)
exp1$TV.post.dev.cost<-abs(7-exp1$TV.rank.cost.post)
exp1$fridge.post.dev.cost<-abs(4-exp1$fridge.rank.cost.post)
exp1$indoor.post.dev.cost<-abs(10-exp1$indoor.rank.cost.post)                   

##AC pre-post signed ranks tests for cost##

rdACc<-wilcox.test(exp1$AC.pre.dev.cost,exp1$AC.post.dev.cost,paired=TRUE,data=exp1)
ACC1<-wilcox.test(exp1$AC.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$AC.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
ACC2<-wilcox.test(exp1$AC.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$AC.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
ACC3<-wilcox.test(exp1$AC.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$AC.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
ACC4<-wilcox.test(exp1$AC.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$AC.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
ACC5<-wilcox.test(exp1$AC.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$AC.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
ACC6<-wilcox.test(exp1$AC.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$AC.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
ACC7<-wilcox.test(exp1$AC.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$AC.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

cAC1.pr<-exp1$AC.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cAC2.pr<-exp1$AC.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cAC3.pr<-exp1$AC.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cAC4.pr<-exp1$AC.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cAC5.pr<-exp1$AC.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cAC6.pr<-exp1$AC.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cAC7.pr<-exp1$AC.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]
cAC1.po<-exp1$AC.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cAC2.po<-exp1$AC.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cAC3.po<-exp1$AC.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cAC4.po<-exp1$AC.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cAC5.po<-exp1$AC.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cAC6.po<-exp1$AC.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cAC7.po<-exp1$AC.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]

cmAC1.pr<-mean(cAC1.pr,na.rm=TRUE)
cmAC1.po<-mean(cAC1.po,na.rm=TRUE)
cmAC2.pr<-mean(cAC2.pr,na.rm=TRUE)
cmAC2.po<-mean(cAC2.po,na.rm=TRUE)
cmAC3.pr<-mean(cAC3.pr,na.rm=TRUE)
cmAC3.po<-mean(cAC3.po,na.rm=TRUE)
cmAC4.pr<-mean(cAC4.pr,na.rm=TRUE)
cmAC4.po<-mean(cAC4.po,na.rm=TRUE)
cmAC5.pr<-mean(cAC5.pr,na.rm=TRUE)
cmAC5.po<-mean(cAC5.po,na.rm=TRUE)
cmAC6.pr<-mean(cAC6.pr,na.rm=TRUE)
cmAC6.po<-mean(cAC6.po,na.rm=TRUE)
cmAC7.pr<-mean(cAC7.pr,na.rm=TRUE)
cmAC7.po<-mean(cAC7.po,na.rm=TRUE)

##Dryer pre-post signed ranks tests for cost##

rddryerc<-wilcox.test(exp1$dryer.pre.dev.cost,exp1$dryer.post.dev.cost,paired=TRUE,data=exp1)
DRC1<-wilcox.test(exp1$dryer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$dryer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
DRC2<-wilcox.test(exp1$dryer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$dryer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
DRC3<-wilcox.test(exp1$dryer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$dryer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
DRC4<-wilcox.test(exp1$dryer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$dryer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
DRC5<-wilcox.test(exp1$dryer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$dryer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
DRC6<-wilcox.test(exp1$dryer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$dryer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
DRC7<-wilcox.test(exp1$dryer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$dryer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

cdry1.pr<-exp1$dryer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cdry2.pr<-exp1$dryer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cdry3.pr<-exp1$dryer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cdry4.pr<-exp1$dryer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cdry5.pr<-exp1$dryer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cdry6.pr<-exp1$dryer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cdry7.pr<-exp1$dryer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]
cdry1.po<-exp1$dryer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cdry2.po<-exp1$dryer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cdry3.po<-exp1$dryer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cdry4.po<-exp1$dryer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cdry5.po<-exp1$dryer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cdry6.po<-exp1$dryer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cdry7.po<-exp1$dryer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]

cmdry1.pr<-mean(cdry1.pr,na.rm=TRUE)
cmdry1.po<-mean(cdry1.po,na.rm=TRUE)
cmdry2.pr<-mean(cdry2.pr,na.rm=TRUE)
cmdry2.po<-mean(cdry2.po,na.rm=TRUE)
cmdry3.pr<-mean(cdry3.pr,na.rm=TRUE)
cmdry3.po<-mean(cdry3.po,na.rm=TRUE)
cmdry4.pr<-mean(cdry4.pr,na.rm=TRUE)
cmdry4.po<-mean(cdry4.po,na.rm=TRUE)
cmdry5.pr<-mean(cdry5.pr,na.rm=TRUE)
cmdry5.po<-mean(cdry5.po,na.rm=TRUE)
cmdry6.pr<-mean(cdry6.pr,na.rm=TRUE)
cmdry6.po<-mean(cdry6.po,na.rm=TRUE)
cmdry7.pr<-mean(cdry7.pr,na.rm=TRUE)
cmdry7.po<-mean(cdry7.po,na.rm=TRUE)

##Oven pre-post signed ranks tests for cost##

rdovenc<-wilcox.test(exp1$oven.pre.dev.cost,exp1$oven.post.dev.cost,paired=TRUE,data=exp1)
OVC1<-wilcox.test(exp1$oven.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$oven.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
OVC2<-wilcox.test(exp1$oven.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$oven.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
OVC3<-wilcox.test(exp1$oven.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$oven.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
OVC4<-wilcox.test(exp1$oven.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$oven.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
OVC5<-wilcox.test(exp1$oven.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$oven.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
OVC6<-wilcox.test(exp1$oven.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$oven.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
OVC7<-wilcox.test(exp1$oven.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$oven.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

cov1.pr<-exp1$oven.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cov2.pr<-exp1$oven.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cov3.pr<-exp1$oven.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cov4.pr<-exp1$oven.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cov5.pr<-exp1$oven.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cov6.pr<-exp1$oven.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cov7.pr<-exp1$oven.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]
cov1.po<-exp1$oven.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cov2.po<-exp1$oven.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cov3.po<-exp1$oven.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cov4.po<-exp1$oven.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cov5.po<-exp1$oven.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cov6.po<-exp1$oven.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cov7.po<-exp1$oven.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]

cmov1.pr<-mean(cov1.pr,na.rm=TRUE)
cmov1.po<-mean(cov1.po,na.rm=TRUE)
cmov2.pr<-mean(cov2.pr,na.rm=TRUE)
cmov2.po<-mean(cov2.po,na.rm=TRUE)
cmov3.pr<-mean(cov3.pr,na.rm=TRUE)
cmov3.po<-mean(cov3.po,na.rm=TRUE)
cmov4.pr<-mean(cov4.pr,na.rm=TRUE)
cmov4.po<-mean(cov4.po,na.rm=TRUE)
cmov5.pr<-mean(cov5.pr,na.rm=TRUE)
cmov5.po<-mean(cov5.po,na.rm=TRUE)
cmov6.pr<-mean(cov6.pr,na.rm=TRUE)
cmov6.po<-mean(cov6.po,na.rm=TRUE)
cmov7.pr<-mean(cov7.pr,na.rm=TRUE)
cmov7.po<-mean(cov7.po,na.rm=TRUE)

##Microwave pre-post signed ranks tests for cost##
rdmicroc<-wilcox.test(exp1$micro.pre.dev.cost,exp1$micro.post.dev.cost,paired=TRUE,data=exp1)
MIC1<-wilcox.test(exp1$micro.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$micro.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
MIC2<-wilcox.test(exp1$micro.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$micro.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
MIC3<-wilcox.test(exp1$micro.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$micro.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
MIC4<-wilcox.test(exp1$micro.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$micro.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
MIC5<-wilcox.test(exp1$micro.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$micro.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
MIC6<-wilcox.test(exp1$micro.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$micro.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
MIC7<-wilcox.test(exp1$micro.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$micro.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

cmi1.pr<-exp1$micro.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cmi2.pr<-exp1$micro.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cmi3.pr<-exp1$micro.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cmi4.pr<-exp1$micro.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cmi5.pr<-exp1$micro.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cmi6.pr<-exp1$micro.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cmi7.pr<-exp1$micro.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]
cmi1.po<-exp1$micro.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cmi2.po<-exp1$micro.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cmi3.po<-exp1$micro.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cmi4.po<-exp1$micro.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cmi5.po<-exp1$micro.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cmi6.po<-exp1$micro.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cmi7.po<-exp1$micro.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]

cmmi1.pr<-mean(cmi1.pr,na.rm=TRUE)
cmmi1.po<-mean(cmi1.po,na.rm=TRUE)
cmmi2.pr<-mean(cmi2.pr,na.rm=TRUE)
cmmi2.po<-mean(cmi2.po,na.rm=TRUE)
cmmi3.pr<-mean(cmi3.pr,na.rm=TRUE)
cmmi3.po<-mean(cmi3.po,na.rm=TRUE)
cmmi4.pr<-mean(cmi4.pr,na.rm=TRUE)
cmmi4.po<-mean(cmi4.po,na.rm=TRUE)
cmmi5.pr<-mean(cmi5.pr,na.rm=TRUE)
cmmi5.po<-mean(cmi5.po,na.rm=TRUE)
cmmi6.pr<-mean(cmi6.pr,na.rm=TRUE)
cmmi6.po<-mean(cmi6.po,na.rm=TRUE)
cmmi7.pr<-mean(cmi7.pr,na.rm=TRUE)
cmmi7.po<-mean(cmi7.po,na.rm=TRUE)

##Heater pre-post signed ranks tests for cost##
#I found a negative effect of total kWh on heater, but gabrielle didn't.  heater were all p<0.10 for AS and kWh.

rdheaterc<-wilcox.test(exp1$heater.pre.dev.cost,exp1$heater.post.dev.cost,paired=TRUE,data=exp1)
HEC1<-wilcox.test(exp1$heater.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$heater.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
HEC2<-wilcox.test(exp1$heater.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$heater.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
HEC3<-wilcox.test(exp1$heater.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$heater.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
HEC4<-wilcox.test(exp1$heater.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$heater.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
HEC5<-wilcox.test(exp1$heater.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$heater.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
HEC6<-wilcox.test(exp1$heater.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$heater.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
HEC7<-wilcox.test(exp1$heater.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$heater.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

che1.pr<-exp1$heater.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
che2.pr<-exp1$heater.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
che3.pr<-exp1$heater.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
che4.pr<-exp1$heater.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
che5.pr<-exp1$heater.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
che6.pr<-exp1$heater.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
che7.pr<-exp1$heater.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]
che1.po<-exp1$heater.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
che2.po<-exp1$heater.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
che3.po<-exp1$heater.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
che4.po<-exp1$heater.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
che5.po<-exp1$heater.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
che6.po<-exp1$heater.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
che7.po<-exp1$heater.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]

cmhe1.pr<-mean(che1.pr,na.rm=TRUE)
cmhe1.po<-mean(che1.po,na.rm=TRUE)
cmhe2.pr<-mean(che2.pr,na.rm=TRUE)
cmhe2.po<-mean(che2.po,na.rm=TRUE)
cmhe3.pr<-mean(che3.pr,na.rm=TRUE)
cmhe3.po<-mean(che3.po,na.rm=TRUE)
cmhe4.pr<-mean(che4.pr,na.rm=TRUE)
cmhe4.po<-mean(che4.po,na.rm=TRUE)
cmhe5.pr<-mean(che5.pr,na.rm=TRUE)
cmhe5.po<-mean(che5.po,na.rm=TRUE)
cmhe6.pr<-mean(che6.pr,na.rm=TRUE)
cmhe6.po<-mean(che6.po,na.rm=TRUE)
cmhe7.pr<-mean(che7.pr,na.rm=TRUE)
cmhe7.po<-mean(che7.po,na.rm=TRUE)

##Washer pre-post signed ranks tests for cost##

rdwasherc<-wilcox.test(exp1$washer.pre.dev.cost,exp1$washer.post.dev.cost,paired=TRUE,data=exp1)
WAC1<-wilcox.test(exp1$washer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$washer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
WAC2<-wilcox.test(exp1$washer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$washer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
WAC3<-wilcox.test(exp1$washer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$washer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
WAC4<-wilcox.test(exp1$washer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$washer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
WAC5<-wilcox.test(exp1$washer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$washer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
WAC6<-wilcox.test(exp1$washer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$washer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
WAC7<-wilcox.test(exp1$washer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$washer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

cwa1.pr<-exp1$washer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cwa2.pr<-exp1$washer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cwa3.pr<-exp1$washer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cwa4.pr<-exp1$washer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cwa5.pr<-exp1$washer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cwa6.pr<-exp1$washer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cwa7.pr<-exp1$washer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]
cwa1.po<-exp1$washer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cwa2.po<-exp1$washer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cwa3.po<-exp1$washer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cwa4.po<-exp1$washer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cwa5.po<-exp1$washer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cwa6.po<-exp1$washer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cwa7.po<-exp1$washer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]

cmwa1.pr<-mean(cwa1.pr,na.rm=TRUE)
cmwa1.po<-mean(cwa1.po,na.rm=TRUE)
cmwa2.pr<-mean(cwa2.pr,na.rm=TRUE)
cmwa2.po<-mean(cwa2.po,na.rm=TRUE)
cmwa3.pr<-mean(cwa3.pr,na.rm=TRUE)
cmwa3.po<-mean(cwa3.po,na.rm=TRUE)
cmwa4.pr<-mean(cwa4.pr,na.rm=TRUE)
cmwa4.po<-mean(cwa4.po,na.rm=TRUE)
cmwa5.pr<-mean(cwa5.pr,na.rm=TRUE)
cmwa5.po<-mean(cwa5.po,na.rm=TRUE)
cmwa6.pr<-mean(cwa6.pr,na.rm=TRUE)
cmwa6.po<-mean(cwa6.po,na.rm=TRUE)
cmwa7.pr<-mean(cwa7.pr,na.rm=TRUE)
cmwa7.po<-mean(cwa7.po,na.rm=TRUE)


##Freezer pre-post signed ranks tests for cost##
##I found no benefits for freezer across any condition, but gabrielle did. 

rdfreezerc<-wilcox.test(exp1$freezer.pre.dev.cost,exp1$freezer.post.dev.cost,paired=TRUE,data=exp1)
FRC1<-wilcox.test(exp1$freezer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$freezer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
FRC2<-wilcox.test(exp1$freezer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$freezer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
FRC3<-wilcox.test(exp1$freezer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$freezer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
FRC4<-wilcox.test(exp1$freezer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$freezer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
FRC5<-wilcox.test(exp1$freezer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$freezer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
FRC6<-wilcox.test(exp1$freezer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$freezer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
FRC7<-wilcox.test(exp1$freezer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$freezer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

cfr1.pr<-exp1$freezer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cfr2.pr<-exp1$freezer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cfr3.pr<-exp1$freezer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cfr4.pr<-exp1$freezer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cfr5.pr<-exp1$freezer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cfr6.pr<-exp1$freezer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cfr7.pr<-exp1$freezer.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]
cfr1.po<-exp1$freezer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cfr2.po<-exp1$freezer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cfr3.po<-exp1$freezer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cfr4.po<-exp1$freezer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cfr5.po<-exp1$freezer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cfr6.po<-exp1$freezer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cfr7.po<-exp1$freezer.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]

cmfr1.pr<-mean(cfr1.pr,na.rm=TRUE)
cmfr1.po<-mean(cfr1.po,na.rm=TRUE)
cmfr2.pr<-mean(cfr2.pr,na.rm=TRUE)
cmfr2.po<-mean(cfr2.po,na.rm=TRUE)
cmfr3.pr<-mean(cfr3.pr,na.rm=TRUE)
cmfr3.po<-mean(cfr3.po,na.rm=TRUE)
cmfr4.pr<-mean(cfr4.pr,na.rm=TRUE)
cmfr4.po<-mean(cfr4.po,na.rm=TRUE)
cmfr5.pr<-mean(cfr5.pr,na.rm=TRUE)
cmfr5.po<-mean(cfr5.po,na.rm=TRUE)
cmfr6.pr<-mean(cfr6.pr,na.rm=TRUE)
cmfr6.po<-mean(cfr6.po,na.rm=TRUE)
cmfr7.pr<-mean(cfr7.pr,na.rm=TRUE)
cmfr7.po<-mean(cfr7.po,na.rm=TRUE)

##TV pre-post signed ranks tests for cost##
#I found no effect of total kWh on TV, Gabrielle did.  I found a benefit for TV from appliance specific \$ and kWh, but gabrielle did not.

rdTVc<-wilcox.test(exp1$TV.pre.dev.cost,exp1$TV.post.dev.cost,paired=TRUE,data=exp1)
TVC1<-wilcox.test(exp1$TV.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$TV.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
TVC2<-wilcox.test(exp1$TV.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$TV.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
TVC3<-wilcox.test(exp1$TV.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$TV.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
TVC4<-wilcox.test(exp1$TV.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$TV.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
TVC5<-wilcox.test(exp1$TV.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$TV.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
TVC6<-wilcox.test(exp1$TV.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$TV.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
TVC7<-wilcox.test(exp1$TV.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$TV.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

cTV1.pr<-exp1$TV.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cTV2.pr<-exp1$TV.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cTV3.pr<-exp1$TV.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cTV4.pr<-exp1$TV.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cTV5.pr<-exp1$TV.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cTV6.pr<-exp1$TV.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cTV7.pr<-exp1$TV.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]
cTV1.po<-exp1$TV.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cTV2.po<-exp1$TV.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cTV3.po<-exp1$TV.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cTV4.po<-exp1$TV.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cTV5.po<-exp1$TV.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cTV6.po<-exp1$TV.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cTV7.po<-exp1$TV.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]

cmTV1.pr<-mean(cTV1.pr,na.rm=TRUE)
cmTV1.po<-mean(cTV1.po,na.rm=TRUE)
cmTV2.pr<-mean(cTV2.pr,na.rm=TRUE)
cmTV2.po<-mean(cTV2.po,na.rm=TRUE)
cmTV3.pr<-mean(cTV3.pr,na.rm=TRUE)
cmTV3.po<-mean(cTV3.po,na.rm=TRUE)
cmTV4.pr<-mean(cTV4.pr,na.rm=TRUE)
cmTV4.po<-mean(cTV4.po,na.rm=TRUE)
cmTV5.pr<-mean(cTV5.pr,na.rm=TRUE)
cmTV5.po<-mean(cTV5.po,na.rm=TRUE)
cmTV6.pr<-mean(cTV6.pr,na.rm=TRUE)
cmTV6.po<-mean(cTV6.po,na.rm=TRUE)
cmTV7.pr<-mean(cTV7.pr,na.rm=TRUE)
cmTV7.po<-mean(cTV7.po,na.rm=TRUE)

##Fridge pre-post signed ranks tests for cost##

rdfridgec<-wilcox.test(exp1$fridge.pre.dev.cost,exp1$fridge.post.dev.cost,paired=TRUE,data=exp1)
FDC1<-wilcox.test(exp1$fridge.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$fridge.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
FDC2<-wilcox.test(exp1$fridge.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$fridge.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
FDC3<-wilcox.test(exp1$fridge.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$fridge.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
FDC4<-wilcox.test(exp1$fridge.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$fridge.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
FDC5<-wilcox.test(exp1$fridge.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$fridge.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
FDC6<-wilcox.test(exp1$fridge.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$fridge.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
FDC7<-wilcox.test(exp1$fridge.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$fridge.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

cFD1.pr<-exp1$FD.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cFD2.pr<-exp1$FD.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cFD3.pr<-exp1$FD.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cFD4.pr<-exp1$FD.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cFD5.pr<-exp1$FD.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cFD6.pr<-exp1$FD.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cFD7.pr<-exp1$FD.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]
cFD1.po<-exp1$FD.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cFD2.po<-exp1$FD.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cFD3.po<-exp1$FD.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cFD4.po<-exp1$FD.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cFD5.po<-exp1$FD.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cFD6.po<-exp1$FD.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cFD7.po<-exp1$FD.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]

cmFD1.pr<-mean(cFD1.pr,na.rm=TRUE)
cmFD1.po<-mean(cFD1.po,na.rm=TRUE)
cmFD2.pr<-mean(cFD2.pr,na.rm=TRUE)
cmFD2.po<-mean(cFD2.po,na.rm=TRUE)
cmFD3.pr<-mean(cFD3.pr,na.rm=TRUE)
cmFD3.po<-mean(cFD3.po,na.rm=TRUE)
cmFD4.pr<-mean(cFD4.pr,na.rm=TRUE)
cmFD4.po<-mean(cFD4.po,na.rm=TRUE)
cmFD5.pr<-mean(cFD5.pr,na.rm=TRUE)
cmFD5.po<-mean(cFD5.po,na.rm=TRUE)
cmFD6.pr<-mean(cFD6.pr,na.rm=TRUE)
cmFD6.po<-mean(cFD6.po,na.rm=TRUE)
cmFD7.pr<-mean(cFD7.pr,na.rm=TRUE)
cmFD7.po<-mean(cFD7.po,na.rm=TRUE)

##Indoor pre-post signed ranks tests for cost##

rdindoorc<-wilcox.test(exp1$indoor.pre.dev.cost,exp1$indoor.post.dev.cost,paired=TRUE,data=exp1)
INC1<-wilcox.test(exp1$indoor.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$indoor.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
INC2<-wilcox.test(exp1$indoor.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$indoor.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
INC3<-wilcox.test(exp1$indoor.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$indoor.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
INC4<-wilcox.test(exp1$indoor.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$indoor.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
INC5<-wilcox.test(exp1$indoor.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$indoor.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
INC6<-wilcox.test(exp1$indoor.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$indoor.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
INC7<-wilcox.test(exp1$indoor.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$indoor.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

cIN1.pr<-exp1$IN.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cIN2.pr<-exp1$IN.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cIN3.pr<-exp1$IN.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cIN4.pr<-exp1$IN.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cIN5.pr<-exp1$IN.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cIN6.pr<-exp1$IN.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cIN7.pr<-exp1$IN.pre.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]
cIN1.po<-exp1$IN.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[1]]
cIN2.po<-exp1$IN.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[2]]
cIN3.po<-exp1$IN.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[3]]
cIN4.po<-exp1$IN.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[4]]
cIN5.po<-exp1$IN.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[5]]
cIN6.po<-exp1$IN.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[6]]
cIN7.po<-exp1$IN.post.dev.cost[exp1$condition==levels(as.factor(exp1$condition))[7]]

cmIN1.pr<-mean(cIN1.pr,na.rm=TRUE)
cmIN1.po<-mean(cIN1.po,na.rm=TRUE)
cmIN2.pr<-mean(cIN2.pr,na.rm=TRUE)
cmIN2.po<-mean(cIN2.po,na.rm=TRUE)
cmIN3.pr<-mean(cIN3.pr,na.rm=TRUE)
cmIN3.po<-mean(cIN3.po,na.rm=TRUE)
cmIN4.pr<-mean(cIN4.pr,na.rm=TRUE)
cmIN4.po<-mean(cIN4.po,na.rm=TRUE)
cmIN5.pr<-mean(cIN5.pr,na.rm=TRUE)
cmIN5.po<-mean(cIN5.po,na.rm=TRUE)
cmIN6.pr<-mean(cIN6.pr,na.rm=TRUE)
cmIN6.po<-mean(cIN6.po,na.rm=TRUE)
cmIN7.pr<-mean(cIN7.pr,na.rm=TRUE)
cmIN7.po<-mean(cIN7.po,na.rm=TRUE)

####
##pre-test use ranks graphs##
gg<-c(0.5,1.5,2.5,3.5,4.5,5.5,6.5,7.5,8.5,9.5,10.5)
xx<-c(0,11)

png(file="preuse1.png",width=1500,height=1600,res=175)
par(mfrow=c(5,2))
hist(exp1$AC.rank.use.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="AC-Pre",ylim=c(0,200))
abline(v=1,col="black",lwd=5,lty=3)
hist(exp1$AC.rank.use.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="AC-Post",ylim=c(0,200))
abline(v=1,col="black",lwd=5,lty=3)
hist(exp1$dryer.rank.use.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Dryer-Pre",ylim=c(0,100))
abline(v=2,col="red",lwd=5,lty=1)
hist(exp1$dryer.rank.use.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Dryer-Post",ylim=c(0,100))
abline(v=2,col="red",lwd=5,lty=1)
hist(exp1$oven.rank.use.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Oven-Pre",ylim=c(0,100))
abline(v=3,col="red",lwd=5,lty=1)
hist(exp1$oven.rank.use.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Oven-Post",ylim=c(0,100))
abline(v=3,col="red",lwd=5,lty=1)
hist(exp1$microwave.rank.use.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Microwave-Pre",ylim=c(0,100))
abline(v=4,col="black",lwd=5,lty=3)
hist(exp1$microwave.rank.use.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Microwave-Post",ylim=c(0,100))
abline(v=4,col="black",lwd=5,lty=3)
hist(exp1$water.heater.rank.use.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Water Heater-Pre",ylim=c(0,100))
abline(v=5,col="red",lwd=5,lty=1)
hist(exp1$water.heater.rank.use.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Water Heater-Post",ylim=c(0,100))
abline(v=5,col="red",lwd=5,lty=1)
dev.off()

png(file="preuse2.png",width=1500,height=1600,res=175)
par(mfrow=c(5,2))
hist(exp1$washer.rank.use.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Washer-Pre",ylim=c(0,100))
abline(v=6,col="red",lwd=5,lty=1)
hist(exp1$washer.rank.use.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Washer-Post",ylim=c(0,100))
abline(v=6,col="red",lwd=5,lty=1)
hist(exp1$freezer.rank.use.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Freezer-Pre",ylim=c(0,100))
abline(v=7,col="red",lwd=5,lty=1)
hist(exp1$freezer.rank.use.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Freezer-Post",ylim=c(0,100))
abline(v=7,col="red",lwd=5,lty=1)
hist(exp1$TV.rank.use.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="TV-Pre",ylim=c(0,100))
abline(v=8,col="red",lwd=5,lty=1)
hist(exp1$TV.rank.use.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="TV-Post",ylim=c(0,100))
abline(v=8,col="red",lwd=5,lty=1)
hist(exp1$fridge.rank.use.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Fridge-Pre",ylim=c(0,100))
abline(v=9,col="black",lwd=5,lty=3)
hist(exp1$fridge.rank.use.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Fridge-Post",ylim=c(0,100))
abline(v=9,col="black",lwd=5,lty=3)
hist(exp1$indoor.rank.use.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="indoor Lights-Pre",ylim=c(0,200))
abline(v=10,col="red",lwd=5,lty=1)
hist(exp1$indoor.rank.use.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="indoor Lightsx-Post",ylim=c(0,200))
abline(v=10,col="red",lwd=5,lty=1)
dev.off()

####
##pre-test cost ranks graphs##
png(file="precost1.png",width=1500,height=1600,res=175)
par(mfrow=c(5,2))
hist(exp1$AC.rank.cost.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="AC-Pre",ylim=c(0,200))
abline(v=1,col="red",lwd=5,lty=1)
hist(exp1$AC.rank.cost.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="AC-Post",ylim=c(0,200))
abline(v=1,col="red",lwd=5,lty=1)
hist(exp1$water.heater.rank.cost.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Water Heater-Pre",ylim=c(0,100))
abline(v=2,col="red",lwd=5,lty=1)
hist(exp1$water.heater.rank.cost.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Water Heater-Post",ylim=c(0,100))
abline(v=2,col="red",lwd=5,lty=1)
hist(exp1$freezer.rank.cost.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Freezer-Pre",ylim=c(0,100))
abline(v=3,col="black",lwd=5,lty=3)
hist(exp1$freezer.rank.cost.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Freezer-Post",ylim=c(0,100))
abline(v=3,col="black",lwd=5,lty=3)
hist(exp1$fridge.rank.cost.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Fridge-Pre",ylim=c(0,100))
abline(v=4,col="red",lwd=5,lty=1)
hist(exp1$fridge.rank.cost.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Fridge-Post",ylim=c(0,100))
abline(v=4,col="red",lwd=5,lty=1)
hist(exp1$dryer.rank.cost.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Dryer-Pre",ylim=c(0,100))
abline(v=5,col="red",lwd=5,lty=1)
hist(exp1$dryer.rank.cost.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Dryer-Post",ylim=c(0,100))
abline(v=5,col="red",lwd=5,lty=1)
dev.off()

png(file="precost2.png",width=1500,height=1600,res=175)
par(mfrow=c(5,2))
hist(exp1$oven.rank.cost.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Oven-Pre",ylim=c(0,100))
abline(v=6,col="red",lwd=5,lty=1)
hist(exp1$oven.rank.cost.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Oven-Post",ylim=c(0,100))
abline(v=6,col="red",lwd=5,lty=1)
hist(exp1$TV.rank.cost.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="TV-Pre",ylim=c(0,100))
abline(v=7,col="red",lwd=5,lty=1)
hist(exp1$TV.rank.cost.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="TV-Post",ylim=c(0,100))
abline(v=7,col="red",lwd=5,lty=1)
hist(exp1$microwave.rank.cost.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Microwave-Pre",ylim=c(0,100))
abline(v=8,col="red",lwd=5,lty=1)
hist(exp1$microwave.rank.cost.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Microwave-Post",ylim=c(0,100))
abline(v=8,col="red",lwd=5,lty=1)
hist(exp1$washer.rank.cost.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Washer-Pre",ylim=c(0,100))
abline(v=9,col="black",lwd=5,lty=3)
hist(exp1$washer.rank.cost.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="Washer-Post",ylim=c(0,100))
abline(v=9,col="black",lwd=5,lty=3)
hist(exp1$indoor.rank.cost.pre,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="indoor Lights-Pre",ylim=c(0,200))
abline(v=10,col="red",lwd=5,lty=1)
hist(exp1$indoor.rank.cost.post,ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main="indoor Lights-Post",ylim=c(0,200))
abline(v=10,col="red",lwd=5,lty=1)
dev.off()

png(file="micro.png",width=1500,height=1600,res=175)
par(mfrow=c(5,2))
hist(exp1$microwave.rank.use.pre[exp1$condition==levels(as.factor(exp1$condition))[1]],ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main=levels(as.factor(exp1$condition))[1],ylim=c(0,20))
abline(v=4,col="red",lwd=5)     
hist(exp1$microwave.rank.use.post[exp1$condition==levels(as.factor(exp1$condition))[1]],ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main=levels(as.factor(exp1$condition))[1],ylim=c(0,20))
abline(v=4,col="red",lwd=5)     
hist(exp1$microwave.rank.use.pre[exp1$condition==levels(as.factor(exp1$condition))[2]],ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main=levels(as.factor(exp1$condition))[2],ylim=c(0,20))
abline(v=4,col="red",lwd=5)     
hist(exp1$microwave.rank.use.post[exp1$condition==levels(as.factor(exp1$condition))[2]],ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main=levels(as.factor(exp1$condition))[2],ylim=c(0,20))
abline(v=4,col="red",lwd=5)     
hist(exp1$microwave.rank.use.pre[exp1$condition==levels(as.factor(exp1$condition))[3]],ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main=levels(as.factor(exp1$condition))[3],ylim=c(0,20))
abline(v=4,col="green",lwd=5)     
hist(exp1$microwave.rank.use.post[exp1$condition==levels(as.factor(exp1$condition))[3]],ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main=levels(as.factor(exp1$condition))[3],ylim=c(0,20))
abline(v=4,col="green",lwd=5)     
hist(exp1$microwave.rank.use.pre[exp1$condition==levels(as.factor(exp1$condition))[4]],ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main=levels(as.factor(exp1$condition))[4],ylim=c(0,20))
abline(v=4,col="red",lwd=5)     
hist(exp1$microwave.rank.use.post[exp1$condition==levels(as.factor(exp1$condition))[4]],ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main=levels(as.factor(exp1$condition))[4],ylim=c(0,20))
abline(v=4,col="red",lwd=5)     
hist(exp1$microwave.rank.use.pre[exp1$condition==levels(as.factor(exp1$condition))[4]],ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main=levels(as.factor(exp1$condition))[4],ylim=c(0,20))
abline(v=4,col="red",lwd=5)     
hist(exp1$microwave.rank.use.post[exp1$condition==levels(as.factor(exp1$condition))[4]],ylab="Frequency",xlab="Rank",xlim=xx,breaks=gg,main=levels(as.factor(exp1$condition))[4],ylim=c(0,20))
abline(v=4,col="red",lwd=5)     
dev.off()

#condition: Display.order..Block.Randomizer.FL_7
#problems: average electricity bill summer; average kwh summer; % of monthly bill for each appliance; pairwise comparisons pre;
ptrunc<-function(x){ifelse(x<0.001,0.001,x)}
##
exp1.x<-exp1[exp1$condition==levels(as.factor(exp1$condition))[1] | exp1$condition==levels(as.factor(exp1$condition))[2] | exp1$condition==levels(as.factor(exp1$condition))[3] | exp1$condition==levels(as.factor(exp1$condition))[4] | exp1$condition==levels(as.factor(exp1$condition))[5] | exp1$condition==levels(as.factor(exp1$condition))[6],]          
exp1.x$AS<-ifelse(exp1.x$condition==levels(as.factor(exp1.x$condition))[1] | exp1.x$condition==levels(as.factor(exp1.x$condition))[3] | exp1.x$condition==levels(as.factor(exp1.x$condition))[4],1,0)
exp1.x$doll<-ifelse(exp1.x$condition==levels(as.factor(exp1.x$condition))[1] | exp1.x$condition==levels(as.factor(exp1.x$condition))[2] | exp1.x$condition==levels(as.factor(exp1.x$condition))[3] | exp1.x$condition==levels(as.factor(exp1.x$condition))[5],1,0)
exp1.x$kWh<-ifelse(exp1.x$condition==levels(as.factor(exp1.x$condition))[1] | exp1.x$condition==levels(as.factor(exp1.x$condition))[2] | exp1.x$condition==levels(as.factor(exp1.x$condition))[4] | exp1.x$condition==levels(as.factor(exp1.x$condition))[6],1,0)
lmAC<-lm((AC.pre.dev.use-AC.post.dev.use)~AS*doll*kWh-1,data=exp1.x)
lmdr<-lm((dryer.pre.dev.use-dryer.post.dev.use)~AS*doll*kWh-1,data=exp1.x)
lmov<-lm((oven.pre.dev.use-oven.post.dev.use)~AS*doll*kWh-1,data=exp1.x)
lmmi<-lm((micro.pre.dev.use-micro.post.dev.use)~AS*doll*kWh-1,data=exp1.x)
lmhe<-lm((heater.pre.dev.use-heater.post.dev.use)~AS*doll*kWh-1,data=exp1.x)
lmwa<-lm((washer.pre.dev.use-washer.post.dev.use)~AS*doll*kWh-1,data=exp1.x)
lmfr<-lm((freezer.pre.dev.use-freezer.post.dev.use)~AS*doll*kWh-1,data=exp1.x)
lmtv<-lm((TV.pre.dev.use-TV.post.dev.use)~AS*doll*kWh-1,data=exp1.x)
lmfri<-lm((fridge.pre.dev.use-fridge.post.dev.use)~AS*doll*kWh-1,data=exp1.x)
lmin<-lm((indoor.pre.dev.use-indoor.post.dev.use)~AS*doll*kWh-1,data=exp1.x)      
##hierarchical model for use##
clmAC<-lm((AC.pre.dev.cost-AC.post.dev.cost)~AS*doll*kWh-1,data=exp1.x)
clmdr<-lm((dryer.pre.dev.cost-dryer.post.dev.cost)~AS*doll*kWh-1,data=exp1.x)
clmov<-lm((oven.pre.dev.cost-oven.post.dev.cost)~AS*doll*kWh-1,data=exp1.x)
clmmi<-lm((micro.pre.dev.cost-micro.post.dev.cost)~AS*doll*kWh-1,data=exp1.x)
clmhe<-lm((heater.pre.dev.cost-heater.post.dev.cost)~AS*doll*kWh-1,data=exp1.x)
clmwa<-lm((washer.pre.dev.cost-washer.post.dev.cost)~AS*doll*kWh-1,data=exp1.x)
clmfr<-lm((freezer.pre.dev.cost-freezer.post.dev.cost)~AS*doll*kWh-1,data=exp1.x)
clmtv<-lm((TV.pre.dev.cost-TV.post.dev.cost)~AS*doll*kWh-1,data=exp1.x)
clmfri<-lm((fridge.pre.dev.cost-fridge.post.dev.cost)~AS*doll*kWh-1,data=exp1.x)
clmin<-lm((indoor.pre.dev.cost-indoor.post.dev.cost)~AS*doll*kWh-1,data=exp1.x)      
##hierarchical model for cost##
@ 

\subsection{Participants}

Participants were bill-paying electricity customers ($N=$\Sexpr{prettyNum(length(na.omit(exp1$age)))#$}) recruited using the Amazon MTurk system.  Forty-two percent were male, most participants were between 22 and 34 years old.  Their income ranged from X to XK per year, with most participants having an income of x per year. The average electricity bill among these customers was \$138/month. 

\subsection{Measures}

Learning from the simulation was measured by assessing participants' rankings of the ten appliances in terms of how much kWh the appliance used in a 10-minute period as well as their rankings of these in terms of their contribution to the monthly bill. For example, participants were asked to ``Imagine each appliance listed below is used for exactly the same amount of time (10 minutes).  Rank the ten appliances below by how much electricity they use from 1 (the most) to 10 (the least).''  The specific text of these questions may be found in Appendix~\ref{app:simmeasures}.  Additional measures and results not discussed for brevity are in Appendix~\ref{app:addmeasures}.

\subsection{Procedure}

Participants completed the ranking questions (pre-test) prior to viewing the IHD simulation. They were then randomized to one of the seven conditions previously mentioned. Interaction with the simulation (and viewing the passive learning information) lasted for as long as they wanted. After interacting with the simulation, they completed the same ranking questions (post-test). Lastly, they completed demographic questions.

\subsection{Results}
<<time,echo=false,results=hide,fig=false>>=
#install.packages("quantreg")
library(quantreg)
exp1.t<-read.csv("simdata1.csv",na.string="",skip=1,header=TRUE,stringsAsFactors=FALSE)
exp1.t$StartDate
exp1.t$EndDate
start.hours<-60*as.numeric(substr(exp1.t$StartDate,12,13))
start.minutes<-as.numeric(substr(exp1.t$StartDate,15,16))
start.seconds<-as.numeric(substr(exp1.t$StartDate,18,19))/60
start.total<-start.hours+start.minutes+start.seconds
stop.hours<-60*as.numeric(substr(exp1.t$EndDate,12,13))
stop.minutes<-as.numeric(substr(exp1.t$EndDate,15,16))
stop.seconds<-as.numeric(substr(exp1.t$EndDate,18,19))/60
stop.total<-stop.hours+stop.minutes+stop.seconds
total.time<-stop.total-start.total
exp1$total.time<-total.time
time.rq<-rq(total.time~condition,data=exp1)
time.rq.sum<-summary.rq(time.rq,se="boot")
@ 

\subsubsection{Level of Interaction}

Participants spent a median of \Sexpr{prettyNum(time.rq.sum$coefficients[1,1])} minutes ($SE=$ \Sexpr{prettyNum(time.rq.sum$coefficients[1,2])} minutes) interacting with the IHD simulation.  Time spent interacting with sim did not differ by conditions (all ts less than \Sexpr{prettyNum(abs(time.rq.sum$coefficients[6,3]))}.

\clearpage
\subsubsection{Appliance Rank Deviations: kWh Use}

Figures~\ref{fig:preuse1} and ~\ref{fig:preuse2} show pre-post differences in rankings of appliances by hourly energy consumption, aggregating across conditions.  As can be seen from the histograms, with red lines indicating the true ranking of the appliance and dashed black lines indicating statistically significant shifts toward the correct ranking at post-test, deviations from the true rank improved for AC, Microwave, and Fridge ($Z=$ \Sexpr{prettyNum(qnorm(1-(rdACu$p.value)/2))}, $p=$ \Sexpr{prettyNum(ptrunc(rdACu$p.value))}; $Z=$ \Sexpr{prettyNum(qnorm(1-(rdmicrou$p.value)/2))}, $p<$ \Sexpr{prettyNum(ptrunc(rdmicrou$p.value))}, and $Z=$ \Sexpr{prettyNum(qnorm(1-(rdfridgeu$p.value)/2))}, $p=$ \Sexpr{prettyNum(ptrunc(rdfridgeu$p.value))}, respectively).  There were no overall differences for the dryer, oven, water heater, washer, freezer, TV, and indoor lights, ($Z=$ \Sexpr{prettyNum(qnorm(1-(rddryeru$p.value)/2))}, $p=$ \Sexpr{prettyNum(rddryeru$p.value)}; $Z=$ \Sexpr{prettyNum(qnorm(1-(rdovenu$p.value)/2))}, $p=$ \Sexpr{prettyNum(rdovenu$p.value)}; $Z=$ \Sexpr{prettyNum(qnorm(1-(rdheateru$p.value)/2))}, $p=$ \Sexpr{prettyNum(rdheateru$p.value)}; $Z=$ \Sexpr{prettyNum(qnorm(1-(rdwasheru$p.value)/2))}, $p=$ \Sexpr{prettyNum(rdwasheru$p.value)}; $Z=$ \Sexpr{prettyNum(qnorm(1-(rdfreezeru$p.value)/2))}, $p=$ \Sexpr{prettyNum(rdfreezeru$p.value)}; $Z=$ \Sexpr{prettyNum(qnorm(1-(rdTVu$p.value)/2))}, $p=$ \Sexpr{prettyNum(rdTVu$p.value)}; and $Z=$ \Sexpr{prettyNum(qnorm(1-(rdindooru$p.value)/2))}, $p=$ \Sexpr{prettyNum(rdindooru$p.value)}, respectively).
 
%$V=$ \Sexpr{prettyNum(rdmicrou$statistic)}
%$V=$ \Sexpr{prettyNum(rdACu$statistic)},
%$V=$ \Sexpr{prettyNum(rdfridgeu$statistic)}
%$V=$ \Sexpr{prettyNum(rddryeru$statistic)}
%$V=$ \Sexpr{prettyNum(rdovenu$statistic)}
%$V=$ \Sexpr{prettyNum(rdheateru$statistic)}
%$V=$ \Sexpr{prettyNum(rdwasheru$statistic)}
%$V=$ \Sexpr{prettyNum(rdfreezeru$statistic)}
%$V=$ \Sexpr{prettyNum(rdTVu$statistic)}
%$V=$ \Sexpr{prettyNum(rdindooru$statistic)}

\begin{figure}[h]
    \centering
\scalebox{1.4}{\includegraphics{preuse1}}
\caption{Histogram of appliance use rankings for each appliance in order of true ranking.  Lines indicate true rank, with black dashed lines indicating statistically significant pre-post differences.}
\label{fig:preuse1}
\end{figure}

\begin{figure}[h]
    \centering
\scalebox{1.4}{\includegraphics{preuse2}}
\caption{Histogram of appliance use rankings for each appliance in order of true ranking.  Lines indicate true rank, with black dashed lines indicating statistically significant pre-post differences.}
\label{fig:preuse2}
\end{figure}

Participants were very accurate in ranking the highest and lowest use appliances, the AC and indoor lights, both before and after interacting with the simulated IHD. Those appliances that participants were initially least sure about (as illustrated by the scattered nature of their initial rankings), the water heater and the oven, did not improve after interacting with the simulated IHD. The most striking improvement was for the microwave, an appliance that participants had certain but incorrect initial beliefs about.

\clearpage
\subsubsection{Appliance Rank Deviations: Cost}

Figures~\ref{fig:precost1} and~\ref{fig:precost2} show pre-post differences in rankings of appliances by monthly contribution to electricity bill, aggregating across conditions.  Results for cost rankings are similar to those for kWh use.  Participants improved from their initial rankings across conditions for the washer and freezer ($Z=$ \Sexpr{prettyNum(qnorm(1-(rdwasherc$p.value)/2))}, $p<$ \Sexpr{prettyNum(ptrunc(rdwasherc$p.value))} and $Z=$ \Sexpr{prettyNum(qnorm(1-(rdfreezerc$p.value)/2))}, $p<$ \Sexpr{prettyNum(ptrunc(rdfreezerc$p.value))}, respectively).  Their rankings at post-test were worse than their pre-test rankings for the oven, water heater, and microwave ($Z=$ \Sexpr{prettyNum(qnorm(1-(rdovenc$p.value)/2))}, $p=$ \Sexpr{prettyNum(rdovenc$p.value)}; $Z=$ \Sexpr{prettyNum(qnorm(1-(rdheaterc$p.value)/2))}, $p=$ \Sexpr{prettyNum(rdheaterc$p.value)}; $Z=$ \Sexpr{prettyNum(qnorm(1-(rdmicroc$p.value)/2))}, $p=$ \Sexpr{prettyNum(ptrunc(rdmicroc$p.value))}, respectively).  For the AC, lights, TV, Dryer, and Fridge, participants were initially quite accurate in their rankings and either remained accurate or slightly improved ($Z=$ \Sexpr{prettyNum(qnorm(1-(rdACc$p.value)/2))}, $p=$ \Sexpr{prettyNum(ptrunc(rdACc$p.value))}; $Z=$ \Sexpr{prettyNum(qnorm(1-(rdindoorc$p.value)/2))}, $p=$ \Sexpr{prettyNum(rdindoorc$p.value)}; $Z=$ \Sexpr{prettyNum(qnorm(1-(rdTVc$p.value)/2))}, $p=$ \Sexpr{prettyNum(rdTVc$p.value)}; $Z=$ \Sexpr{prettyNum(qnorm(1-(rddryerc$p.value)/2))}, $p=$ \Sexpr{prettyNum(rddryerc$p.value)}; $Z=$ \Sexpr{prettyNum(qnorm(1-(rdfridgec$p.value)/2))}, $p=$ \Sexpr{prettyNum(ptrunc(rdfridgec$p.value))}, respectively). 

%$V=$ \Sexpr{prettyNum(rdwasherc$statistic)}
%$V=$ \Sexpr{prettyNum(rdfreezerc$statistic)}
%$V=$ \Sexpr{prettyNum(rdACc$statistic)}
%$V=$ \Sexpr{prettyNum(rdmicroc$statistic)}
%$V=$ \Sexpr{prettyNum(rdfridgec$statistic)}
%$V=$ \Sexpr{prettyNum(rddryerc$statistic)}
%$V=$ \Sexpr{prettyNum(rdovenc$statistic)}
%$V=$ \Sexpr{prettyNum(rdheaterc$statistic)}
%$V=$ \Sexpr{prettyNum(rdTVc$statistic)}
%$V=$ \Sexpr{prettyNum(rdindoorc$statistic)}

\begin{figure}[h]
    \centering
\scalebox{1.4}{\includegraphics{precost1}}
\caption{Histogram of appliance monthly cost rankings for each appliance in order of true ranking.  Lines indicate true rank, with black dashed lines indicating statistically significant pre-post differences.}
\label{fig:precost1}
\end{figure}

\begin{figure}[h]
    \centering
\scalebox{1.4}{\includegraphics{precost2}}
\caption{Histogram of appliance monthly cost rankings for each appliance in order of true ranking.  Lines indicate true rank, with black dashed lines indicating statistically significant pre-post differences.}
\label{fig:precost2}
\end{figure}

Contrary to the findings for kWh use, the post-simulation rankings for the microwave were more inaccurate than the pre-simulation rankings. However, unlike their precise but incorrect initial beliefs about how many kWhs it used in 10 minutes, participants had imprecise beliefs about how much the microwave would cost them in a month. Participants learned that the microwave used much more energy than they expected, but they seemed to incorrectly extrapolate this greater energy use to monthly cost, not taking into account how infrequently the microwave is used compared to other appliances. More detailed analyses can be found in Appendix~\ref{app:simcost}.

\subsubsection{Appliance Rank Deviations by Specific Condition}

Following the aggregate analysis, we then looked at how the rankings of each appliance changed for each treatment condition individually.  As can be seen in Table~\ref{tab:raneff} and X, a random effects model \cite{gelman2007data,gelman2010arm} with different intercepts for each appliance examines the effect of each treatment condition aggregating across appliances, weighting them appropriately.  Overall, more improvements in ranking accuracy occur in conditions in which kWh rather than \$ units or appliance-specific information are provided.  Moreover, participants appear unable to extrapolate from monthly cost feedback to monthly kWh use. The appliance specific information does not appear to contribute to improvements in ranking accuracy. Detailed analyses can be found in Appendix~\ref{app:kWhrank}.

<<randomeffects,results=hide,echo=false,fig=false>>=
exp1.pass<-exp1[exp1$condition=="Passive Learning",]
f<-c()
frame.pass<-data.frame()
for(i in 1:length(levels(as.factor(exp1.pass$ResponseID)))){
f.ac<-c(exp1.pass$ResponseID[i],exp1.pass$AC.pre.dev.use[i],exp1.pass$AC.post.dev.use[i],"AC")
f.dryer<-c(exp1.pass$ResponseID[i],exp1.pass$dryer.pre.dev.use[i],exp1.pass$dryer.post.dev.use[i],"Dryer")
f.micro<-c(exp1.pass$ResponseID[i],exp1.pass$micro.pre.dev.use[i],exp1.pass$micro.post.dev.use[i],"Microwave")
f.oven<-c(exp1.pass$ResponseID[i],exp1.pass$oven.pre.dev.use[i],exp1.pass$oven.post.dev.use[i],"Oven")
f.heater<-c(exp1.pass$ResponseID[i],exp1.pass$heater.pre.dev.use[i],exp1.pass$heater.post.dev.use[i],"Heater")
f.tv<-c(exp1.pass$ResponseID[i],exp1.pass$TV.pre.dev.use[i],exp1.pass$TV.post.dev.use[i],"TV")
f.fridge<-c(exp1.pass$ResponseID[i],exp1.pass$fridge.pre.dev.use[i],exp1.pass$fridge.post.dev.use[i],"Fridge")
f.lights<-c(exp1.pass$ResponseID[i],exp1.pass$indoor.pre.dev.use[i],exp1.pass$indoor.post.dev.use[i],"Lights")
f.washer<-c(exp1.pass$ResponseID[i],exp1.pass$washer.pre.dev.use[i],exp1.pass$washer.post.dev.use[i],"Washer")
f.freezer<-c(exp1.pass$ResponseID[i],exp1.pass$freezer.pre.dev.use[i],exp1.pass$freezer.post.dev.use[i],"Freezer")
g<-rbind(f.ac,f.dryer)
g<-rbind(g,f.micro)
g<-rbind(g,f.oven)
g<-rbind(g,f.heater)
g<-rbind(g,f.tv)
g<-rbind(g,f.fridge)
g<-rbind(g,f.lights)
g<-rbind(g,f.washer)
g<-rbind(g,f.freezer)
frame.pass<-rbind(frame.pass,g)
}
colnames(frame.pass)<-c("ID","pre","post","Appliance")
frame.pass$pre<-as.numeric(frame.pass$pre)
frame.pass$post<-as.numeric(frame.pass$post)

f<-c()
frame<-data.frame()
for(i in 1:length(levels(as.factor(exp1.x$ResponseID)))){
f.ac<-c(exp1.x$ResponseID[i],exp1.x$AC.pre.dev.use[i],exp1.x$AC.post.dev.use[i],"AC",exp1.x$doll[i],exp1.x$kWh[i],exp1.x$AS[i])
f.dryer<-c(exp1.x$ResponseID[i],exp1.x$dryer.pre.dev.use[i],exp1.x$dryer.post.dev.use[i],"Dryer",exp1.x$doll[i],exp1.x$kWh[i],exp1.x$AS[i])
f.micro<-c(exp1.x$ResponseID[i],exp1.x$micro.pre.dev.use[i],exp1.x$micro.post.dev.use[i],"Microwave",exp1.x$doll[i],exp1.x$kWh[i],exp1.x$AS[i])
f.oven<-c(exp1.x$ResponseID[i],exp1.x$oven.pre.dev.use[i],exp1.x$oven.post.dev.use[i],"Oven",exp1.x$doll[i],exp1.x$kWh[i],exp1.x$AS[i])
f.heater<-c(exp1.x$ResponseID[i],exp1.x$heater.pre.dev.use[i],exp1.x$heater.post.dev.use[i],"Heater",exp1.x$doll[i],exp1.x$kWh[i],exp1.x$AS[i])
f.tv<-c(exp1.x$ResponseID[i],exp1.x$TV.pre.dev.use[i],exp1.x$TV.post.dev.use[i],"TV",exp1.x$doll[i],exp1.x$kWh[i],exp1.x$AS[i])
f.fridge<-c(exp1.x$ResponseID[i],exp1.x$fridge.pre.dev.use[i],exp1.x$fridge.post.dev.use[i],"Fridge",exp1.x$doll[i],exp1.x$kWh[i],exp1.x$AS[i])
f.lights<-c(exp1.x$ResponseID[i],exp1.x$indoor.pre.dev.use[i],exp1.x$indoor.post.dev.use[i],"Lights",exp1.x$doll[i],exp1.x$kWh[i],exp1.x$AS[i])
f.washer<-c(exp1.x$ResponseID[i],exp1.x$washer.pre.dev.use[i],exp1.x$washer.post.dev.use[i],"Washer",exp1.x$doll[i],exp1.x$kWh[i],exp1.x$AS[i])
f.freezer<-c(exp1.x$ResponseID[i],exp1.x$freezer.pre.dev.use[i],exp1.x$freezer.post.dev.use[i],"Freezer",exp1.x$doll[i],exp1.x$kWh[i],exp1.x$AS[i])
g<-rbind(f.ac,f.dryer)
g<-rbind(g,f.micro)
g<-rbind(g,f.oven)
g<-rbind(g,f.heater)
g<-rbind(g,f.tv)
g<-rbind(g,f.fridge)
g<-rbind(g,f.lights)
g<-rbind(g,f.washer)
g<-rbind(g,f.freezer)
frame<-rbind(frame,g)
}
colnames(frame)<-c("ID","pre","post","Appliance","Dollars","kWh","AS")
frame$pre<-as.numeric(frame$pre)
frame$post<-as.numeric(frame$post)
frame$Dollars<-factor(frame$Dollars,levels=c(0,1))
frame$kWh<-factor(frame$kWh,levels=c(0,1))
frame$AS<-factor(frame$AS,levels=c(0,1))

library(arm)
null<-lmer(post~pre+(1|Appliance),data=frame)
all1<-lmer((pre-post)~kWh+AS+Dollars+(1|Appliance)+(1|ID),data=frame)
kwh1<-lmer((pre-post)~1+(1|Appliance)+(1|ID),data=frame[frame$kWh==1,])
pass1<-lmer((pre-post)~1+(1|Appliance)+(1|ID),data=frame.pass)
@ 

\begin{table}[h]
\centering
  \caption{Hierarchical linear model with varying-intercepts for appliance and participant.}
  \label{tab:raneff}
  \begin{tabular}{c c c c c}
    Condition & Improvement & t-Statistic & p-value \\ \hline
    kWh & \Sexpr{prettyNum(fixef(all1)[2])} & \Sexpr{prettyNum(fixef(all1)[2]/sqrt(diag(vcov(all1))[2]))} & \Sexpr{prettyNum(dt(fixef(all1)[2]/sqrt(diag(vcov(all1))[2]),1486))} \\ 
    AS & \Sexpr{prettyNum(fixef(all1)[3])} & \Sexpr{prettyNum(fixef(all1)[3]/sqrt(diag(vcov(all1))[3]))} & \Sexpr{prettyNum(dt(fixef(all1)[3]/sqrt(diag(vcov(all1))[3]),1486))} \\  
    \$ & \Sexpr{prettyNum(fixef(all1)[4])} & \Sexpr{prettyNum(fixef(all1)[4]/sqrt(diag(vcov(all1))[4]))} & \Sexpr{prettyNum(dt(fixef(all1)[4]/sqrt(diag(vcov(all1))[4]),1486))} \\ \hline
    Passive & \Sexpr{prettyNum(fixef(pass1)[1])} & \Sexpr{prettyNum(fixef(pass1)[1]/sqrt(diag(vcov(pass1))[1]))} & \Sexpr{prettyNum(dt(fixef(pass1)[1]/sqrt(diag(vcov(pass1))[1]),1486))} \\ \hline 
\end{tabular} 
\end{table}

\subsection{Discussion}

Participants were able to learn how much energy their appliances used from the simulation. Learning was particularly pronounced for appliances that they had preconceived but incorrect notions about, such as the microwave, which used much more energy than they expected. Learning was more modest or non-existent for appliances that they knew little about (the water heater and the oven) beforehand. Participants also demonstrated high accuracy in identifying which appliances used the most (air conditioner) and least (indoor lights) out of the ten.

Although the effects were small, participants learned more about how much energy the appliances used when provided with kWh feedback, regardless of whether they were also provided feedback in dollars or by appliance. Thus, participants may have difficulty focusing their attention when provided with a large tabular display that provides kWh information on each appliance. This type of `information overload' has been found elsewhere, especially with respect to the well-established literature on working-memory (cite).  This literature suggests that people have the capacity to consider roughly three `chunks' of information at one time. The ineffectiveness of dollar feedback indicates that participants had difficulty translating feedback provided in monetary units to energy units. These two results are important because people overwhelmingly prefer bill-to-date (in dollars) and appliance-specific feedback, but we find no evidence of their effectiveness in learning.

In contrast to participants' ability to learn the energy usage of each appliance, participants were largely unable to learn how much each appliance cost in a month. The finding that participants who became more accurate in their knowledge of how much energy the microwave used, but less accurate in their knowledge of how much it costs per month provides a window into the learning process. It seems as though participants were basing their monthly cost estimates on their energy ranking estimates, failing to take into account how often each appliance is used. This explains the opposing effects for the microwave, as learning that it uses more energy would also lead one to overestimate how much it costs per month. This finding suggests that people need projected monthly costs for their appliances, as they have difficulty extrapolating from current cost and energy use to monthly cost. We found that these projected monthly costs are a feature that participants find desirable, but it is not at the top of their list (it ranked fourth below bill-to-date, appliance-specific feedback, and daily projections).

Perhaps the most surprising finding from the experiment was the consistent success of the passive learning condition. This condition merely provided participants with the answers to the knowledge test, which they were able to remember and subsequently regurgitate. In the real world it is much cheaper and more feasible to just provide a flyer with information about appliance-specific energy consumption, monthly cost, and how people should adjust their behavior to save energy. This cheap and easy solution may not make consumers more knowledgeable about their own electricity consumption, but may induce as much or more behavioral change as an IHD, using a much simpler method. However, the external validity of this finding is questionable for two reasons. First, participants likely did not develop the working knowledge of their appliances needed to extrapolate to new appliances, and thus any benefit from deeper comprehension and associated feelings of mastery and control over ones environment will not emerge. Second, people routinely receive flyers and inserts, which they dispose of without viewing, so this method of providing information may not be useful in the real world because there is saturation in this medium.

\section{General Discussion}
-- a paragraph of set-up, reminding readers of the motivation, why you chose the task, and how you innovated in with it.  
-- a reprise of the results, including your contributions to the literature.  
-- a paragraph on limits and needed research
-- 2-3 paragraphs (max) on practical implications  if these results hold (after noting the differences between the experiments and the world).

Maybe we can say something about how important the extremes (particularly on the heavy energy use side are).  Perhaps we can also say something about how much getting the ranking right for the appliances in the middle matters.

We need to be clear about what we mean by behavior change.  I think what our results allow us to say is that we are testing the configuration of attributes that leads to:
enhanced understanding about the relative costs and energy use of appliances
enhanced understanding of the most effective ways to save energy.

Paragraph of set-up, reminding readers of the motivation, why you chose the task, and how you innovated in with it. – a reprise of the results, including your contributions to the literature. – a paragraph on limits and needed research – 2-3 paragraphs (max) on practical implications if these results hold (after noting the differences between the experiments and the world).

Overall, the results suggest that people need to be provided with only an aggregate summary of their current kWh consumption, a projected aggregate or disaggregate monthly bill, or a flyer that will not be discarded that provides key information.

People seem to have the extremes down. The ones they have stronger beliefs about they seem to learn and the ones theyâre initially confused on, they donât learn.

Two explanations: 1) Familiarity frequency of use * look at data and see if their familiarity matches with our results 2) Familiarity via utility/campaigns * look at the correlation in rank by function (food functions, clothes function etc) Heuristic is same function and same use. It would make you really wrong with the dryer or washer if you are right about one and not the other yet you link them together. 3) Mental model of how it works

\subsection{Future directions}: mental models of `diffuse' appliances

First, we look at several appliances that participants seemed to disagree in their rankings about: oven, water heater, washer, freezer, and fridge.  We also look at mental models of the microwave, that was consistently underestimated in rank.  We want to know how they think these appliances work, and why they had such disagreement on their rankings (maybe do group discussions?). 

Maybe people didn't get the point of it, it was boring or whatever.  Thus, this is a test of effectiveness rather than efficacy, which is good, as these are problems that will probably happen in the real world (baruch's quote). 

Follow-up studies and stuff for discussion section:

a) Intervention where we get them to explain how it works to get them to learn. Control intervention that says look at same appliance or think about same appliance without “explaining”

b) Provide them with new appliances and if they really understood the previous list of appliances, they should be able to extrapolate. If they don’t they should look just like passive learning people. Passive learning people should be really crappy on extrapolating from one condition to the new appliances. 

\subsection{Conclusion}


\appendix
\clearpage
\section{Supplemental Review of Customer Preferences}
\label{app:pref}

\subsection{Units}
The information displayed on an IHD can come in different units, such as current cost (\$), cost/day, power (W or kW), energy (kWh), and carbon emissions (CO2 tons).  In general, people prefer the cost of electricity above all other possible ways to display electricity use \cite{karjalainen2011consumer,anderson2009exploring}.  This is consistent with customers wanting simple information in units that they already understand.  A number of studies have found that people want to see their costs either as current rate of expenditures (in \$/day) or cumulative cost in \$ per billing period \cite{darby2010smart,anderson2009exploring}.  A few studies have found that people want to be able to switch from the cost of electricity to kWh when looking at the IHD, with one study finding that people prefer energy (kWh) over power (W or kW) \cite{karjalainen2011consumer}.

\subsection{Time Aggregation}
The displayed information can come in increments of years, quarters, months, weeks, days, hours, real-time or other periodic cycles.  Unlike preferences for units, there appears to be no consensus regarding people's preferences for how time should be aggregated.  Some, for example, prefer to see their electricity consumption on an hourly basis \cite{anderson2009exploring}, while others prefer to see it on a quarterly basis, compared to some reference point like the previous quarter \cite{roberts2004consumer}.  Still others prefer to see their electricity use displayed as daily load curves \cite{ueno2005effectiveness} rather than ten-day curves \cite{allen2006effects}.  However, while there is no unanimous preference for time-period, people generally want to be able to switch time periods with the press of a single button \cite{darby2010smart,anderson2009exploring}.  While people are used to getting information monthly, when they receive their bill, more frequent information may be helpful \cite{wilhite1995measured}.

\subsection{Physical Aggregation}
While we know of no research on whether people prefer electricity use information by room, by specific household member, or for the whole house, a recent study found that people strongly prefer appliance-specific information \cite{karjalainen2011consumer}, and they want to see this information in dollar units \cite{anderson2009exploring}.  

\subsection{Comparators}
Comparisons typically examined have been to one-self (historically), to other customers (social comparisons), or to targets (goals).  The most frequent finding is that people want to compare their current use to their own use at some point in the past \cite{roberts2004consumer,karjalainen2011consumer}.  Moreover, people want to compare their personal electricity use to a self-set goal or target \cite{darby2010smart,anderson2009exploring}.  In contrast, nearly all people express a strong rejection of social comparisons \cite{roberts2004consumer,anderson2009exploring,paetz2011shifting}, wherein they see their electricity use compared to some other group of customers, such as their neighbors.  Indeed, there is little evidence suggesting social comparisons motivate people to reduce their household electricity use \cite{fischer2008feedback}. 

\subsection{Format}
Displayed information can be formatted as a chart, picture, table, numerically, as text, or as a combination of audio and visual feedback.  Like time aggregation, there is a lot of variability in preferences for format. 

\emph{Visual analogs}.  Several studies have shown that people like visual-analog `speedometers' that show cumulative use or rate of use \cite{anderson2009exploring}.  This partially explains why many people rate the GEO minim, which uses this type of format in its display, highly (\cite{darby2010smart}; green energy options; \url{http://www.greenenergyoptions.co.uk/}).

\emph{Graphical displays}.  Customers usually prefer graphical displays in the form of bar graphs, although some people like bell-curves as well \cite{roberts2003towards}.  Karjalainen \cite{karjalainen2011consumer} developed eight paper-based IHD prototypes and evaluated them using preference assessments and think-aloud protocols.  People understood bar charts, pie charts, and numerical tables easily.  However, the researchers found that a tabular display was not only, ``instantly understood by everyone,'' but also preferred the most, as 7/15 participants ranked it then best of the eight prototypes.  

\emph{Text, diagrams and colors}.  There is some evidence that combinations of texts and diagrams may enhance each other \cite{roberts2003towards}.  However, gimmicky pictures (e.g., stacks of coins, different sized houses) are strongly disliked \cite{roberts2004consumer}.  Paetz \cite{paetz2011shifting} found that colors were useful for showing price information, as one participant wrote in their diary that ``With a quick glance I was able to check the current price. It is quick and easy to comprehend with the three colors.''

\subsection{Other}
A few other attributes have been proposed as potentially proving effective in enhancing people's understanding and use of information:  (a) Motivation, including incentives, tariffs, and goals; (b) interactivity, allowing the user to determine what appears on the display dynamically; (c) grouping of information by room, type of appliance, subgroups, or hierarchically; (d) oracles, advisors, or intelligent tutors, that provide people tips, suggestions, answers, or even act on behalf of the customer.

\clearpage
\section{Survey Measures}
\label{app:surveymeasures}

\begin{quote}
Here is a list of information that might appear on an in-home display.  Please rate each type of information in terms of how much you would like to have it on the display.
\end{quote}

\begin{enumerate}
\item How many dollars you've spent on electricity so far this month. [IA1]
\item How much electricity each of your appliances uses. [IA2]
\item How many dollars you will spend each day, if you keep using electricity at the same rate. [IA3]
\item How much electricity you will use by the end of the month, based on how much you've used so far. [IA4]
\item How many Kilowatt hours of electricity you've used so far this month. [IA5]
\item The daily price of electricity---if you are on a program where electricity prices differ each day. [IA6]
\item The time of day you use the most electricity. [IA7]
\item When you used the most electricity over the last month. [IA8]
\item How well you're meeting a goal that you've set for how much electricity you will use. [IA9]
\item How ``green'' your electricity use is each month. [IA10]
\item How your electricity use compares with that of a neighbor or a similar household. [IA11]
\end{enumerate}

All special features were also rated on a 1-5 scale (not at all to extremely) or ``I don't know.''  Code for each measure is in brackets (e.g.,[SF1]).  They were given the following instructions:

\begin{quote}
Below is an extensive list of special features that can be added to the in-home display. Please rate each feature on how much you would like it to be on a display.
\end{quote}

\begin{enumerate}  
\item You can set the in-home display to automatically control appliances in your home, such as your air conditioner. [SF1]
\item It allows you to pre-set a spending limit and alerts you when you are approaching or passed this limit. [SF2]
\item It warns you if a power outage (blackout) is likely. [SF3]
\item It helps you set a goal of how much electricity to conserve and tracks whether you're approaching your goal. [SF4]
\item It has graphics that change color when electricity prices are more or less expensive than usual. [SF5]
\item You can set it to provide you with personal alerts about how much electricity you've used or how much money you’ve spent on electricity that month. [SF6]
\item It provides movies and educational tutorials that explain how to use the display. [SF7]
\item It has a screensaver that is pleasing to look at when you are not using the display to access information. [SF8]
\item It provides you with audio feedback as well as visual feedback (for example, it might talk to you or sound alarms). [SF9]
\item You can upload and view digital photographs on the display. [SF10]
\item You can read blogs and access the internet on the display screen. [SF11]
\item You can play games on the display. [SF12]
\item You can access social networking sites (e.g., Facebook) on the display. [SF13]
\end{enumerate}

\subsection{Special Features}

The rankings of the special features can be seen in Table~\ref{tab:feat}.  None of the top six features (automatic appliance control, spending alerts, blackout notice, set and track goals, changing colors, and monthly alerts) differed in important ways.  The second tier of features were educational tutorials, audio and visual feedback, and a pleasing screensaver.  The least preferred features were ones that would make the display multi-purpose, such as photographs, use of blogs and the internet, playing games, and Facebook.  These were all rated significantly lower than the lowest of the second tier of features (audio/visual feedback), $V=3489$, $p<0.001$.

\begin{table}[h]
  \caption{Special Feature Preferences}
  \label{tab:feat}
  \centering
  \begin{tabular}{l c c}
   Item & Mean & SD \\ \hline
   Automatic appliance control [SF1] & 3.81 & 1.09 \\
   Spending alerts [SF2] & 3.75 & 1.04 \\
   Blackout notice [SF3] & 3.75 & 1.15 \\
   Set and track goals [SF4] & 3.71 & 1.08 \\
   Changing color [SF5] & 3.63 & 1.12 \\
   Monthly alerts [SF6] & 3.60 & 1.07 \\
   Tutorials [SF7] & 2.97 & 1.29 \\
   Screensaver [SF9] & 2.83 &  1.34 \\
   Audio/Visual [SF10] & 2.89 & 1.34 \\
   Photographs [SF10] & 2.16 & 1.37 \\
   Blogs and Internet [SF11] & 2.07 & 1.33 \\
   Games [SF12] & 1.84 & 1.21 \\
   Social Networking [SF13] & 1.84 & 1.25 \\ \hline
   \end{tabular}
  \end{table}

\subsection{Retail Preferences}
Participants then rated 19 retail devices on aesthetics and expected usefulness.  To make the task easier, participants only rated a subsample of five of them.  All ratings were also rated on a 1-5 scale (not at all to extremely) or ``I don't know.''  Code for each measure is in brackets (e.g.,[RL1]).  They were given the following instructions: 

\begin{quote}
Here are five commercially available in-home displays . Please rate each one in terms of (a) how much you like how it looks and (b) how useful it would be to you.
How much do you like the way this display looks?
\end{quote}

Here is an example of the description (the rest are in appendix 1):
\begin{quote}
Eco-Meter (Click image below for larger picture)
\begin{itemize}
\item Easy-to-read with large numbers and LED back-lighting.     
\item Designed to sit on your counter-top.
\item Uses LED lights to alert you about important events (for ex., a period of high cost electricity).
\item Displays the cost of the electricity you’ve been using.
\item >More about Eco-Meter
\end{itemize}
\end{quote}

Participants then rated 19 retail devices on aesthetics and expected usefulness.  To make the task eaier, participants only rated a subsample of five of them.  All ratings were also rated on a 1-5 scale (not at all to extremely) or ``I don't know.''  Code for each measure is in brackets (e.g. [RL1]).  They were given the following instructions: 

\begin{quote}
Here are five commercially available in-home displays . Please rate each one in terms of (a) how much you like how it looks and (b) how useful it would be to you. 
Overall, how useful would this display be?
\end{quote}

The devices were the EnergyHub HomeBase, OpenFrame, Control4 EC100, HEC, Silverstat 7, 

\begin{table}
  \caption{Retail Preferences for Aesthetics and Usefulness}
  \centering
  \begin{tabular}{l c c c c c}
 & \multicolumn{2}{c}{Aesthetics} & \multicolumn{2}{c}{Usefulness} &\\
 Item & Mean & SD & Mean & SD & N \\ \hline
 HomeBase & 4.17 & 0.79 & 3.67 & 0.92 & 35 \\
 OpenFrame & 3.96 & 1.06 & 3.64 & 1.15 & 27 \\
 EC100 & 3.90 & 0.84 & 3.76 & 0.96 & 101 \\
 HEC & 3.77 & 1.12 & 3.66 & 1.26 & 31 \\
 Silverstat 7 & 3.77 & 1.09 & 3.30 & 0.95 & 13 \\
 PowerCost Monitor & 3.76 & 0.93 & 3.82 & 1.07 & 37 \\
 Insight & 3.69 & 1.00 & 4.15 & 0.77 & 29 \\
 Envi & 3.61 & 0.99 & 3.85 & 0.77 & 28 \\
 Elite & 3.54 & 0.93 & 3.82 & 0.80 & 24 \\
 Cent-A-Meter & 3.49 & 0.77 & 3.50 & 0.77 & 37 \\
 Powertab & 3.43 & 1.27 & 3.14 & 0.69 & 7 \\
 TED5000 & 3.37 & 1.05 & 3.70 & 0.89 & 113 \\
 GEO Minim &  3.33 & 1.18 & 3.26 & 1.20 & 30 \\
 E2 & 3.30 & 0.67 & 3.18 & 1.17 & 10 \\
 Ratesaver & 3.25 & 0.75 & 3.25 & 0.75 & 12\\
 Intellifocus & 2.97 & 0.98 & 3.11 & 0.93 & 86 \\
 Eco-Meter & 2.96 & 1.00 & 3.12 & 0.97 & 26 \\
 Emu & 2.92 & 1.00 & 2.83 & 0.72 & 12 \\
 Eco-Eye & 2.50 & 1.27 & 2.80 & 1.14 & 10 \\ \hline
 \end{tabular}
  \end{table}



Participants then created their own display by selecting what features they wanted from those listed above.  Their constructions were almost identical to their ratings, so we omit them here.  They also responded to the following questions about the display they created

\begin{quote}
  \begin{itemize}
   \item  How much would you like to have the in-home display you created on the previous page? (not at all, slightly, moderately, strongly, extremely) 
   \item How effective do you think that in-home display would help you to reduce your electricity use?'' (very ineffective, ineffective, somewhat ineffective, neither effective nor ineffective, somewhat effective, effective, very effective)
   \item How often would you look at the in-home display you created? (Never, less than once a month, once a month, 2-3 times a month, once a week, 2-3 times a week, dialy).
   \item How much do you think you would save, in dollars, on your monthly electricity bill if you had the display you created on the previous page? 
\end{itemize}
  \end{quote}

They liked the display they created, on average expected it to like it strongly (M=3.96, SD=0.84, N=138), they thought it would be effective (M=5.89, SD=1.09; N=139), expected to look at it on average 2-3 times a week (M=6.11, SD=1.1, N=138), expected significant monthly savings on their billa (M=\$36.2, SD=\$40.1, N=106; 95\% CI [28, 44]).

The average willingness to pay for an IHD was \$154 (95\%CI [2,306]) (SD=870, N=131), average WTP for IHD on a recurring monthly bill was \$18 (SD=39, N=121), average expected savings on the monthly will was \$25 (SD=29, N=131), the average monthly electricity bill was \$106 (SD=81, N=106), and they reported using 893 kWh on average per month (SD=1134, N=20), but most either didn't answer the question or reported they didn't know. 	

\clearpage
\section{Sim. Measures}
\label{app:simmeasures}

The electricity knowledge tests measured six main dependent variables to capture learning from the simulation.  The first measure assessed participants' rankings of the ten appliances in terms of how much kWh the appliance used in a 10-minute period:

\begin{quote}
[jack] What was the text for the use?
\end{quote}

Next, participants ranked the ten appliances by their contribution to the monthly bill:

\begin{quote} 
Imagine the average summer monthly cost for a family of two adults and two children for each appliance listed below.  Rank the ten appliances below by how much they would contribute to the monthly bill from 1 (the most) to 10 (the least).
\end{quote} 

The third question asked them to identify the correct unit of electrical energy:

\begin{quote}
  What are the units of electrical energy called?
\begin{itemize}
\item Kilowatt (kW)
\item Kilowatt-hours (kWh)
\item British Thermal Units (BTU)
\item Volts (V)
\item Horsepower (HP)
\end{itemize}
\end{quote}

Participants were then asked to identify the correct method of calculating energy:

\begin{quote}
Which of the following is equal to the amount of energy consumed by an electrical appliance? 
\begin{itemize}
\item Power rating multiplied by the cost of electricity.
\item Power rating added to the cost of electricity.
\item Power rating multiplied by the time it's used.
\item Power rating divided by the time it's used.
\item Power rating added to the time it's used.
\end{itemize}
\end{quote}

Participants then estimated the cost of a kWh (we used \$0.13 as an approximation):

\begin{quote}
``The cost of a kWh varies a little from state to state and from hour to hour.  What is the average cost of a kWh for a residential U.S. customer?''
\end{quote}

Finally, participants estimated the monthly kWh use for households.

\begin{quote}
About how many kWh does a typical U.S. household use in a summer month?
\end{quote}

\clearpage
\section{Additional Measures and Results}
\label{app:addmeasures}
<<energycalcs,results=hide,echo=false,fig=false>>=
##Energy Calculations
exp1$energy.calc.score.pre<-ifelse(exp1$energy.calc.pre=="Power rating multiplied by the time it's used",1,0)
exp1$energy.calc.score.post<-ifelse(exp1$energy.calc.post=="Power rating multiplied by the time it's used",1,0)
#glm<-multinom(energy.calc.score.post-energy.calc.score.pre~condition-1,data=exp1)
##

exp1$energy.calc.score.change<-exp1$energy.calc.score.post-exp1$energy.calc.score.pre

a11<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change>0 & exp1$condition==levels(as.factor(exp1$condition))[1]]))
a12<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change==0 & exp1$condition==levels(as.factor(exp1$condition))[1]]))
a13<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change<0 & exp1$condition==levels(as.factor(exp1$condition))[1]]))

a21<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change>0 & exp1$condition==levels(as.factor(exp1$condition))[2]]))
a22<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change==0 & exp1$condition==levels(as.factor(exp1$condition))[2]]))
a23<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change<0 & exp1$condition==levels(as.factor(exp1$condition))[2]]))

a31<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change>0 & exp1$condition==levels(as.factor(exp1$condition))[3]]))
a32<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change==0 & exp1$condition==levels(as.factor(exp1$condition))[3]]))
a33<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change<0 & exp1$condition==levels(as.factor(exp1$condition))[3]]))

a41<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change>0 & exp1$condition==levels(as.factor(exp1$condition))[4]]))
a42<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change==0 & exp1$condition==levels(as.factor(exp1$condition))[4]]))
a43<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change<0 & exp1$condition==levels(as.factor(exp1$condition))[4]]))

a51<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change>0 & exp1$condition==levels(as.factor(exp1$condition))[5]]))
a52<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change==0 & exp1$condition==levels(as.factor(exp1$condition))[5]]))
a53<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change<0 & exp1$condition==levels(as.factor(exp1$condition))[5]]))

a61<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change>0 & exp1$condition==levels(as.factor(exp1$condition))[6]]))
a62<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change==0 & exp1$condition==levels(as.factor(exp1$condition))[6]]))
a63<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change<0 & exp1$condition==levels(as.factor(exp1$condition))[6]]))

a71<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change>0 & exp1$condition==levels(as.factor(exp1$condition))[7]]))
a72<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change==0 & exp1$condition==levels(as.factor(exp1$condition))[7]]))
a73<-length(na.omit(exp1$energy.calc.score.change[exp1$energy.calc.score.change<0 & exp1$condition==levels(as.factor(exp1$condition))[7]]))

a1<-sum(a11,a21,a31,a41,a51,a61,a71)
a2<-sum(a12,a22,a32,a42,a52,a62,a72)
a3<-sum(a13,a23,a33,a43,a53,a63,a73)
a1.se<-100*sd(rbeta(10000,a1,a2+a3))
a3.se<-100*sd(rbeta(10000,a3,a2+a1))
t<-t1<-(a1-a3)/((a1.se+a3.se)/2)

#install.packages("MCMCpack")
library(MCMCpack)
a11.se<-100*sd(rbeta(10000,a11,a12+a13))
a13.se<-100*sd(rbeta(10000,a13,a11+a12))
t1<-(a11-a13)/((a11.se+a13.se)/2)

a21.se<-100*sd(rbeta(10000,a21,a22+a23))
a23.se<-100*sd(rbeta(10000,a23,a21+a22))
t2<-(a21-a23)/((a21.se+a23.se)/2)

a31.se<-100*sd(rbeta(10000,a31,a32+a33))
a33.se<-100*sd(rbeta(10000,a33,a31+a32))
t3<-(a31-a33)/((a31.se+a33.se)/2)

a41.se<-100*sd(rbeta(10000,a41,a42+a43))
a43.se<-100*sd(rbeta(10000,a43,a41+a42))
t4<-(a41-a43)/((a41.se+a43.se)/2)

a51.se<-100*sd(rbeta(10000,a51,a52+a53))
a53.se<-100*sd(rbeta(10000,a53,a51+a52))
t5<-(a51-a53)/((a51.se+a53.se)/2)

a61.se<-100*sd(rbeta(10000,a61,a62+a63))
a63.se<-100*sd(rbeta(10000,a63,a61+a62))
t6<-(a61-a63)/((a61.se+a63.se)/2)

a71.se<-100*sd(rbeta(10000,a71,a72+a73))
a73.se<-100*sd(rbeta(10000,a73,a71+a72))
t7<-(a71-a73)/((a71.se+a73.se)/2)

@ 

As seen in Table~\ref{tab:encalc}, participants were not more able to infer how energy is calculated in any of the treatments.  Most (\Sexpr{prettyNum(a2)}) participants did not change their answer.  Some (\Sexpr{prettyNum(a1)}) changed from the wrong answer to the right answer, whereas others changed from the right answer to the wrong answer (\Sexpr{prettyNum(a3)}). There was no detectable variation by condition, but aggregating across all conditions there was improvement.

\begin{table}[h]
  \centering
  \begin{tabular}{c c c c c}
Condition                & Correct & No Change & Incorrect & t-value \\ \hline
 \$ and kWh by appliance & \Sexpr{a11} & \Sexpr{a12} & \Sexpr{a13} & \Sexpr{prettyNum(t1)} \\
 \$ and kWh aggregate & \Sexpr{a21} & \Sexpr{a22} & \Sexpr{a23} & \Sexpr{prettyNum(t2)} \\
\$ by appliance & \Sexpr{a31} & \Sexpr{a32} & \Sexpr{a33} & \Sexpr{prettyNum(t3)} \\
 kWh by appliance & \Sexpr{a41} & \Sexpr{a42} & \Sexpr{a43} & \Sexpr{prettyNum(t4)} \\
Aggregate \$ only & \Sexpr{a51} & \Sexpr{a52} & \Sexpr{a53} & \Sexpr{prettyNum(t5)} \\
Aggregate kWh only & \Sexpr{a61} & \Sexpr{a62} & \Sexpr{a63} & \Sexpr{prettyNum(t6)} \\
Passive Learning & \Sexpr{a71} & \Sexpr{a72} & \Sexpr{a73} & \Sexpr{prettyNum(t7)} \\ \hline
Total & \Sexpr{a1} & \Sexpr{a2} & \Sexpr{a3} & \Sexpr{prettyNum(t)} \\ \hline
\end{tabular}
  \caption{Identification of how energy is calculated by condition.  T-values comparing proportion switched to correct versus incorrect for each condition are based on posterior simulations from a beta(0,0) prior.}
\label{tab:encalc}
\end{table}

\subsubsection{Units of Energy}
<<units,echo=false,results=hide,fig=false>>=
##Electricity units#
exp1$electricity.units.score.pre<-ifelse(exp1$electricity.units.pre=="Kilowatt-hours (kWh)",1,0)
exp1$electricity.units.score.post<-ifelse(exp1$electricity.units.post=="Kilowatt-hours (kWh)",1,0)
##
exp1$electricity.units.score.change<-exp1$electricity.units.score.post-exp1$electricity.units.score.pre
nc<-length(na.omit(exp1$ResponseID[exp1$electricity.units.score.change==0]))
right<-length(na.omit(exp1$ResponseID[exp1$electricity.units.score.change==1]))
wrong<-length(na.omit(exp1$ResponseID[exp1$electricity.units.score.change==-1]))

glm1<-glm(electricity.units.score.change~condition,data=exp1[exp1$electricity.units.score.change>=0,],family=binomial(link="logit"))
glm1<-glm(electricity.units.score.post~condition+electricity.units.score.pre-1,data=exp1[exp1$electricity.units.score.change>=0,],family=binomial(link="logit"))
glm2<-glm(abs(electricity.units.score.change)~condition-1,data=exp1[exp1$electricity.units.score.change<=0,],family=binomial(link="logit"))

a11<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change>0 & exp1$condition==levels(as.factor(exp1$condition))[1]]))
a12<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change==0 & exp1$condition==levels(as.factor(exp1$condition))[1]]))
a13<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change<0 & exp1$condition==levels(as.factor(exp1$condition))[1]]))

a21<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change>0 & exp1$condition==levels(as.factor(exp1$condition))[2]]))
a22<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change==0 & exp1$condition==levels(as.factor(exp1$condition))[2]]))
a23<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change<0 & exp1$condition==levels(as.factor(exp1$condition))[2]]))

a31<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change>0 & exp1$condition==levels(as.factor(exp1$condition))[3]]))
a32<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change==0 & exp1$condition==levels(as.factor(exp1$condition))[3]]))
a33<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change<0 & exp1$condition==levels(as.factor(exp1$condition))[3]]))

a41<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change>0 & exp1$condition==levels(as.factor(exp1$condition))[4]]))
a42<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change==0 & exp1$condition==levels(as.factor(exp1$condition))[4]]))
a43<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change<0 & exp1$condition==levels(as.factor(exp1$condition))[4]]))

a51<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change>0 & exp1$condition==levels(as.factor(exp1$condition))[5]]))
a52<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change==0 & exp1$condition==levels(as.factor(exp1$condition))[5]]))
a53<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change<0 & exp1$condition==levels(as.factor(exp1$condition))[5]]))

a61<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change>0 & exp1$condition==levels(as.factor(exp1$condition))[6]]))
a62<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change==0 & exp1$condition==levels(as.factor(exp1$condition))[6]]))
a63<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change<0 & exp1$condition==levels(as.factor(exp1$condition))[6]]))

a71<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change>0 & exp1$condition==levels(as.factor(exp1$condition))[7]]))
a72<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change==0 & exp1$condition==levels(as.factor(exp1$condition))[7]]))
a73<-length(na.omit(exp1$electricity.units.score.change[exp1$electricity.units.score.change<0 & exp1$condition==levels(as.factor(exp1$condition))[7]]))

a1<-sum(a11,a21,a31,a41,a51,a61,a71)
a2<-sum(a12,a22,a32,a42,a52,a62,a72)
a3<-sum(a13,a23,a33,a43,a53,a63,a73)
a1.se<-100*sd(rbeta(10000,a1,a2+a3))
a3.se<-100*sd(rbeta(10000,a3,a2+a1))
t<-t1<-(a1-a3)/((a1.se+a3.se)/2)

#install.packages("MCMCpack")
library(MCMCpack)
a11.se<-100*sd(rbeta(10000,a11,a12+a13))
a13.se<-100*sd(rbeta(10000,a13,a11+a12))
t1<-(a11-a13)/((a11.se+a13.se)/2)

a21.se<-100*sd(rbeta(10000,a21,a22+a23))
a23.se<-100*sd(rbeta(10000,a23,a21+a22))
t2<-(a21-a23)/((a21.se+a23.se)/2)

a31.se<-100*sd(rbeta(10000,a31,a32+a33))
a33.se<-100*sd(rbeta(10000,a33,a31+a32))
t3<-(a31-a33)/((a31.se+a33.se)/2)

a41.se<-100*sd(rbeta(10000,a41,a42+a43))
a43.se<-100*sd(rbeta(10000,a43,a41+a42))
t4<-(a41-a43)/((a41.se+a43.se)/2)

a51.se<-100*sd(rbeta(10000,a51,a52+a53))
a53.se<-100*sd(rbeta(10000,a53,a51+a52))
t5<-(a51-a53)/((a51.se+a53.se)/2)

a61.se<-100*sd(rbeta(10000,a61,a62+a63))
a63.se<-100*sd(rbeta(10000,a63,a61+a62))
t6<-(a61-a63)/((a61.se+a63.se)/2)

a71.se<-100*sd(rbeta(10000,a71,a72+a73))
a73.se<-100*sd(rbeta(10000,a73,a71+a72))
t7<-(a71-a73)/((a71.se+a73.se)/2)
@ 

As seen in Table ~\ref{tab:enunits}, in all conditions participants were more likely to change their answer from an incorrect unit of energy to the correct one (kWh) than from correct to incorrect.  However, none of the differences were statistically significant.  Aggregating across all conditions, participants were more likely to change their answer in the correct direction.  However, there is no control group of participants who did not receive any feedback (i.e., merely completed the pre and post questionnaires), to compare them against.  Thus, this positive change could be an effect of learning from the questions, that frequently asked about kWh, rather than the feedback.

\begin{table}[h]
  \centering
  \begin{tabular}{c c c c c}
Condition                & Correct & No Change & Incorrect & t-value \\ \hline
 \$ and kWh by appliance & \Sexpr{a11} & \Sexpr{a12} & \Sexpr{a13} & \Sexpr{prettyNum(t1)} \\
 \$ and kWh aggregate & \Sexpr{a21} & \Sexpr{a22} & \Sexpr{a23} & \Sexpr{prettyNum(t2)} \\
\$ by appliance & \Sexpr{a31} & \Sexpr{a32} & \Sexpr{a33} & \Sexpr{prettyNum(t3)} \\
 kWh by appliance & \Sexpr{a41} & \Sexpr{a42} & \Sexpr{a43} & \Sexpr{prettyNum(t4)} \\
Aggregate \$ only & \Sexpr{a51} & \Sexpr{a52} & \Sexpr{a53} & \Sexpr{prettyNum(t5)} \\
Aggregate kWh only & \Sexpr{a61} & \Sexpr{a62} & \Sexpr{a63} & \Sexpr{prettyNum(t6)} \\
Passive Learning & \Sexpr{a71} & \Sexpr{a72} & \Sexpr{a73} & \Sexpr{prettyNum(t7)} \\ \hline
Total & \Sexpr{a1} & \Sexpr{a2} & \Sexpr{a3} & \Sexpr{prettyNum(t)} \\ \hline
\end{tabular}
  \caption{Identification of the correct unit of energy (kWh) by condition.  T-values comparing proportion switched to correct versus incorrect for each condition are based on posterior simulations from a beta(0,0) prior.}
\label{tab:enunits}
\end{table}

\subsubsection{Cost of a kWh}
<<kwhcost,echo=false,results=hide,fig=false>>=
library(quantreg)
#kwh.use.pre
exp1$kwh.cost.pre.cents.fix<-as.numeric(exp1$kwh.cost.pre.cents)
exp1$kwh.cost.pre.cents.fix<-ifelse(exp1$kwh.cost.pre.cents.fix>1,exp1$kwh.cost.pre.cents.fix/100,exp1$kwh.cost.pre.cents.fix)
exp1$kwh.cost.pre.doll.fix<-as.numeric(exp1$kwh.cost.pre.doll)

exp1$kwh.cost.pre<-exp1$kwh.cost.pre.cents.fix+exp1$kwh.cost.pre.doll.fix
exp1$kwh.cost.pre.dev<-abs(exp1$kwh.cost.pre-0.13)

exp1$kwh.cost.post.cents.fix<-as.numeric(exp1$kwh.cost.post.cents)
exp1$kwh.cost.post.cents.fix<-ifelse(exp1$kwh.cost.post.cents.fix>1,exp1$kwh.cost.post.cents.fix/100,exp1$kwh.cost.post.cents.fix)
exp1$kwh.cost.post.doll.fix<-as.numeric(exp1$kwh.cost.post.doll)

exp1$kwh.cost.post<-exp1$kwh.cost.post.cents.fix+exp1$kwh.cost.post.doll.fix
exp1$kwh.cost.post.dev<-abs(exp1$kwh.cost.post-0.13)
lm1<-lm(kwh.cost.pre.dev-kwh.cost.post.dev~condition,data=exp1)

png(file="kwhcost.png",width=1500,height=1600,res=175)
#xx<-c(0:1000)
gg<-c(0,1,2,3,4,5,6,7,8,9,10)
#xlim=xx,breaks=gg
par(mfrow=c(5,2))
hist(exp1$kwh.cost.pre[exp1$kwh.cost.pre<10 & exp1$condition==levels(as.factor(exp1$condition))[1]],xlim=c(0,10),ylab="Frequency",breaks=gg,xlab="kWh Cost",main=levels(as.factor(exp1$condition))[1],ylim=c(0,20))
abline(v=0.13,col="red",lwd=5)     
hist(exp1$kwh.cost.post[exp1$kwh.cost.pre<10 & exp1$condition==levels(as.factor(exp1$condition))[1]],xlim=c(0,10),ylab="Frequency",breaks=gg,xlab="kWh Cost",main=levels(as.factor(exp1$condition))[1],ylim=c(0,20))
abline(v=0.13,col="red",lwd=5)     
dev.off()
a11<-median(na.omit(exp1$kwh.cost.pre[exp1$condition==levels(as.factor(exp1$condition))[1]]))
a12<-median(na.omit(exp1$kwh.cost.post[exp1$condition==levels(as.factor(exp1$condition))[1]]))
a21<-median(na.omit(exp1$kwh.cost.pre[exp1$condition==levels(as.factor(exp1$condition))[2]]))
a22<-median(na.omit(exp1$kwh.cost.post[exp1$condition==levels(as.factor(exp1$condition))[2]]))
a31<-median(na.omit(exp1$kwh.cost.pre[exp1$condition==levels(as.factor(exp1$condition))[3]]))
a32<-median(na.omit(exp1$kwh.cost.post[exp1$condition==levels(as.factor(exp1$condition))[3]]))
a41<-median(na.omit(exp1$kwh.cost.pre[exp1$condition==levels(as.factor(exp1$condition))[4]]))
a42<-median(na.omit(exp1$kwh.cost.post[exp1$condition==levels(as.factor(exp1$condition))[4]]))
a51<-median(na.omit(exp1$kwh.cost.pre[exp1$condition==levels(as.factor(exp1$condition))[5]]))
a52<-median(na.omit(exp1$kwh.cost.post[exp1$condition==levels(as.factor(exp1$condition))[5]]))
a61<-median(na.omit(exp1$kwh.cost.pre[exp1$condition==levels(as.factor(exp1$condition))[6]]))
a62<-median(na.omit(exp1$kwh.cost.post[exp1$condition==levels(as.factor(exp1$condition))[6]]))
a71<-median(na.omit(exp1$kwh.cost.pre[exp1$condition==levels(as.factor(exp1$condition))[7]]))
a72<-median(na.omit(exp1$kwh.cost.post[exp1$condition==levels(as.factor(exp1$condition))[7]]))
a1<-median(na.omit(exp1$kwh.cost.pre))
a2<-median(na.omit(exp1$kwh.cost.post))

t1<-wilcox.test(exp1$kwh.cost.post.dev[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$kwh.cost.pre.dev[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
t2<-wilcox.test(exp1$kwh.cost.post.dev[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$kwh.cost.pre.dev[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
t3<-wilcox.test(exp1$kwh.cost.post.dev[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$kwh.cost.pre.dev[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
t4<-wilcox.test(exp1$kwh.cost.post.dev[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$kwh.cost.pre.dev[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
t5<-wilcox.test(exp1$kwh.cost.post.dev[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$kwh.cost.pre.dev[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
t6<-wilcox.test(exp1$kwh.cost.post.dev[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$kwh.cost.pre.dev[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
t7<-wilcox.test(exp1$kwh.cost.post.dev[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$kwh.cost.pre.dev[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)
t<-wilcox.test(exp1$kwh.cost.post.dev,exp1$kwh.cost.pre.dev,paired=TRUE)
@ 

As seen in Table~\ref{tab:kwhcost}, participants in all groups were closer to the true kWh cost of \$0.13 in the post-treatment period, except for those who received \$ and kWh aggregate feedback.  The only statistically significant improvement was for participants in the passive learning condition, the median of which got the true kWh cost exactly correct. 

\begin{table}[h]
  \centering
  \begin{tabular}{c c c c c}
                         & \multicolumn{2}{c}{Median kWh Cost} & \\
Condition                &  Pre        &    Post     & V & p-value \\ \hline
 \$ and kWh by appliance & \Sexpr{a11} & \Sexpr{a12} & \Sexpr{prettyNum(t1$statistic)} & \Sexpr{prettyNum(t1$p.value)} \\
 \$ and kWh aggregate    & \Sexpr{a21} & \Sexpr{a22} & \Sexpr{prettyNum(t2$statistic)} & \Sexpr{prettyNum(t2$p.value)} \\ 
\$ by appliance          & \Sexpr{a31} & \Sexpr{a32} & \Sexpr{prettyNum(t3$statistic)} & \Sexpr{prettyNum(t3$p.value)} \\
 kWh by appliance        & \Sexpr{a41} & \Sexpr{a42} & \Sexpr{prettyNum(t4$statistic)} & \Sexpr{prettyNum(t4$p.value)} \\
Aggregate \$ only        & \Sexpr{a51} & \Sexpr{a52} & \Sexpr{prettyNum(t5$statistic)} & \Sexpr{prettyNum(t5$p.value)} \\
Aggregate kWh only       & \Sexpr{a61} & \Sexpr{a62} & \Sexpr{prettyNum(t6$statistic)} & \Sexpr{prettyNum(t6$p.value)} \\
Passive Learning         & \Sexpr{a71} & \Sexpr{a72} & \Sexpr{prettyNum(t7$statistic)} & \Sexpr{prettyNum(t7$p.value)} \\ \hline
Total                    & \Sexpr{a1}  & \Sexpr{a2}  & \Sexpr{prettyNum(t$statistic)} & \Sexpr{prettyNum(t$p.value)} \\ \hline
\end{tabular}
\caption{Median estimates of the cost of a kWh before and after treatment.  T-values are from Wilcoxon paired rank deviations from the true values.}
\label{tab:kwhcost}
\end{table}

\subsubsection{Monthly kWh}
<<monthlykwh,echo=false,results=hide,fig=false>>=
#kwh.use.pre
exp1$kwh.use.pre.fix<-as.numeric(exp1$kwh.use.pre)
exp1$kwh.use.post.fix<-as.numeric(exp1$kwh.use.post)
exp1$kwh.pre.dev<-abs(exp1$kwh.use.pre.fix-958)
exp1$kwh.post.dev<-abs(exp1$kwh.use.post.fix-958)

a11<-median(na.omit(exp1$kwh.use.pre.fix[exp1$condition==levels(as.factor(exp1$condition))[1]]))
a12<-median(na.omit(exp1$kwh.use.post.fix[exp1$condition==levels(as.factor(exp1$condition))[1]]))
a21<-median(na.omit(exp1$kwh.use.pre.fix[exp1$condition==levels(as.factor(exp1$condition))[2]]))
a22<-median(na.omit(exp1$kwh.use.post.fix[exp1$condition==levels(as.factor(exp1$condition))[2]]))
a31<-median(na.omit(exp1$kwh.use.pre.fix[exp1$condition==levels(as.factor(exp1$condition))[3]]))
a32<-median(na.omit(exp1$kwh.use.post.fix[exp1$condition==levels(as.factor(exp1$condition))[3]]))
a41<-median(na.omit(exp1$kwh.use.pre.fix[exp1$condition==levels(as.factor(exp1$condition))[4]]))
a42<-median(na.omit(exp1$kwh.use.post.fix[exp1$condition==levels(as.factor(exp1$condition))[4]]))
a51<-median(na.omit(exp1$kwh.use.pre.fix[exp1$condition==levels(as.factor(exp1$condition))[5]]))
a52<-median(na.omit(exp1$kwh.use.post.fix[exp1$condition==levels(as.factor(exp1$condition))[5]]))
a61<-median(na.omit(exp1$kwh.use.pre.fix[exp1$condition==levels(as.factor(exp1$condition))[6]]))
a62<-median(na.omit(exp1$kwh.use.post.fix[exp1$condition==levels(as.factor(exp1$condition))[6]]))
a71<-median(na.omit(exp1$kwh.use.pre.fix[exp1$condition==levels(as.factor(exp1$condition))[7]]))
a72<-median(na.omit(exp1$kwh.use.post.fix[exp1$condition==levels(as.factor(exp1$condition))[7]]))

a1<-median(na.omit(exp1$kwh.use.pre))
a2<-median(na.omit(exp1$kwh.use.post))

t1<-wilcox.test(exp1$kwh.post.dev[exp1$condition==levels(as.factor(exp1$condition))[1]],exp1$kwh.pre.dev[exp1$condition==levels(as.factor(exp1$condition))[1]],paired=TRUE)
t2<-wilcox.test(exp1$kwh.post.dev[exp1$condition==levels(as.factor(exp1$condition))[2]],exp1$kwh.pre.dev[exp1$condition==levels(as.factor(exp1$condition))[2]],paired=TRUE)
t3<-wilcox.test(exp1$kwh.post.dev[exp1$condition==levels(as.factor(exp1$condition))[3]],exp1$kwh.pre.dev[exp1$condition==levels(as.factor(exp1$condition))[3]],paired=TRUE)
t4<-wilcox.test(exp1$kwh.post.dev[exp1$condition==levels(as.factor(exp1$condition))[4]],exp1$kwh.pre.dev[exp1$condition==levels(as.factor(exp1$condition))[4]],paired=TRUE)
t5<-wilcox.test(exp1$kwh.post.dev[exp1$condition==levels(as.factor(exp1$condition))[5]],exp1$kwh.pre.dev[exp1$condition==levels(as.factor(exp1$condition))[5]],paired=TRUE)
t6<-wilcox.test(exp1$kwh.post.dev[exp1$condition==levels(as.factor(exp1$condition))[6]],exp1$kwh.pre.dev[exp1$condition==levels(as.factor(exp1$condition))[6]],paired=TRUE)
t7<-wilcox.test(exp1$kwh.post.dev[exp1$condition==levels(as.factor(exp1$condition))[7]],exp1$kwh.pre.dev[exp1$condition==levels(as.factor(exp1$condition))[7]],paired=TRUE)

t<-wilcox.test(exp1$kwh.post.dev,exp1$kwh.pre.dev,paired=TRUE)
@ 

Table~\ref{tab:monthlykwh} shows estimates of monthly kWh use.

\begin{table}[h]
  \centering
  \begin{tabular}{c c c c c}
                         & \multicolumn{2}{c}{Median kWh Use} & \\
Condition                &  Pre        &    Post     & V & p-value \\ \hline
 \$ and kWh by appliance & \Sexpr{a11} & \Sexpr{a12} & \Sexpr{prettyNum(t1$statistic)} & \Sexpr{prettyNum(t1$p.value)} \\
 \$ and kWh aggregate    & \Sexpr{a21} & \Sexpr{a22} & \Sexpr{prettyNum(t2$statistic)} & \Sexpr{prettyNum(t2$p.value)} \\ 
\$ by appliance          & \Sexpr{a31} & \Sexpr{a32} & \Sexpr{prettyNum(t3$statistic)} & \Sexpr{prettyNum(t3$p.value)} \\
 kWh by appliance        & \Sexpr{a41} & \Sexpr{a42} & \Sexpr{prettyNum(t4$statistic)} & \Sexpr{prettyNum(t4$p.value)} \\
Aggregate \$ only        & \Sexpr{a51} & \Sexpr{a52} & \Sexpr{prettyNum(t5$statistic)} & \Sexpr{prettyNum(t5$p.value)} \\
Aggregate kWh only       & \Sexpr{a61} & \Sexpr{a62} & \Sexpr{prettyNum(t6$statistic)} & \Sexpr{prettyNum(t6$p.value)} \\
Passive Learning         & \Sexpr{a71} & \Sexpr{a72} & \Sexpr{prettyNum(t7$statistic)} & \Sexpr{prettyNum(t7$p.value)} \\ \hline
Total                    & \Sexpr{a1}  & \Sexpr{a2}  & \Sexpr{prettyNum(t$statistic)} & \Sexpr{prettyNum(t$p.value)} \\ \hline
\end{tabular}
\caption{Median estimates monthly kWh use before and after treatment.  T-values are from Wilcoxon paired rank deviations from the true values}
\label{tab:monthlykwh}
\end{table}

\clearpage
\section{Detailed kWh rank analyses}
\label{app:kWhrank}
Table~\ref{tab:userank} summarizes the effect of each treatment condition on each appliance.  AC benefited for kWh by appliance for understanding use, $V=$ \Sexpr{prettyNum(ACU4$statistic)}, $p=$ \Sexpr{prettyNum(ACU4$p.value)}, or total only with kWh use, $V=$ \Sexpr{prettyNum(ACU6$statistic)}, $p=$ \Sexpr{prettyNum(ACU6$p.value)}.  Dryer slightly benefitted with total kWh and \$, but not otherwise $V=$ \Sexpr{prettyNum(DRU2$statistic)}, $p=$ \Sexpr{prettyNum(DRU2$p.value)}.  Microwave differed in kWh by appliance $V=$ \Sexpr{prettyNum(MIU4$statistic)}, $p=$ \Sexpr{prettyNum(MIU4$p.value)}, total \$ only $V=$ \Sexpr{prettyNum(MIU5$statistic)}, $p=$ \Sexpr{prettyNum(MIU5$p.value)}, total kWh only $V=$ \Sexpr{prettyNum(MIU6$statistic)}, $p=$ \Sexpr{prettyNum(MIU6$p.value)}, and passive learning $V=$ \Sexpr{prettyNum(MIU7$statistic)}, $p=$ \Sexpr{prettyNum(MIU7$p.value)}.  Heater benefited from kWh by appliance $V=$ \Sexpr{prettyNum(HEU4$statistic)}, $p=$ \Sexpr{prettyNum(HEU4$p.value)}, and passive learning $V=$ \Sexpr{prettyNum(HEU7$statistic)}, $p=$ \Sexpr{prettyNum(HEU7$p.value)}.  TV benefitted from both kWh and \$ by appliance $V=$ \Sexpr{prettyNum(TVU1$statistic)}, $p=$ \Sexpr{prettyNum(TVU1$p.value)}, and passive learning $V=$ \Sexpr{prettyNum(TVU7$statistic)}, $p=$ \Sexpr{prettyNum(TVU7$p.value)}.  The refrigerator benefitted from total kWh only $V=$ \Sexpr{prettyNum(FDU6$statistic)}, $p=$ \Sexpr{prettyNum(FDU6$p.value)}, and passive learning $V=$ \Sexpr{prettyNum(FDU7$statistic)}, $p=$ \Sexpr{prettyNum(FDU7$p.value)}.  Indoor lighting was helped only by passive learning $V=$ \Sexpr{prettyNum(INU7$statistic)}, $p=$ \Sexpr{prettyNum(INU7$p.value)}.The washing machine did not benefit in any condition.  There was no learning across any condition for the freezer.

\begin{table}[h]
  \centering
\scalebox{0.8}{
  \begin{tabular}{c c c c c c c c c}
 Appliance & \$, kWh, AS & \$, kWh & \$, AS & kWh, AS & \$ & kWh & Passive & Total \\ \hline
AC            &  &  &  & +&  & +&  & 2 \\
Dryer         &  & +&  &  &  &  &  & 1 \\
Microwave     &  &  &  & +& +& +& +& 4 \\
Oven          &  &  &  &  & -&  &  & (-1) \\
Water Heater  &  &  &  & +&  & -& +& 2 (-1) \\
TV            & +&  &  &  &  &  & +& 2 \\ 
Fridge        &  &  &  &  &  & +& +& 2 \\ 
Indoor Lights &  &  &  &  &  &  & +& 1 \\ 
Washer        &  &  &  &  &  &  &  & 0 \\ 
Freezer       &  &  &  &  &  &  &  & 0 \\ \hline
Total         & 1& 1& 0& 3& 1 (-1) & 3 (-1) & 5 &  \\ \hline  
  \end{tabular}}
  \caption{Helped (+) or harmed (-) rankings of energy use for each appliance. AS is appliance specific feedback.}
\label{tab:userank}
 \end{table}

\begin{table}
  \caption{Air Conditioner}
  \begin{tabular}{c c c c c}
    Condition & Pre Rank Dev & Post Rank Dev & Statistic & p-value \\ \hline
    \$ only   & \Sexpr{prettyNum(mAC5.pr)} & \Sexpr{prettyNum(mAC5.po)} & \Sexpr{prettyNum(ACU5$statistic)} & \Sexpr{prettyNum(ACU5$p.value)} \\
    kWh only   & \Sexpr{prettyNum(mAC6.pr)} & \Sexpr{prettyNum(mAC6.po)} & \Sexpr{prettyNum(ACU6$statistic)} & \Sexpr{prettyNum(ACU6$p.value)} \\
    \$ and kWh & \Sexpr{prettyNum(mAC2.pr)} & \Sexpr{prettyNum(mAC2.po)} & \Sexpr{prettyNum(ACU2$statistic)} & \Sexpr{prettyNum(ACU2$p.value)} \\
    \$ app.   & \Sexpr{prettyNum(mAC3.pr)} & \Sexpr{prettyNum(mAC3.po)} & \Sexpr{prettyNum(ACU3$statistic)} & \Sexpr{prettyNum(ACU3$p.value)} \\
    kWh app.   & \Sexpr{prettyNum(mAC4.pr)} & \Sexpr{prettyNum(mAC4.po)} & \Sexpr{prettyNum(ACU4$statistic)} & \Sexpr{prettyNum(ACU4$p.value)} \\
    \$ and kWh app.   & \Sexpr{prettyNum(mAC1.pr)} & \Sexpr{prettyNum(mAC1.po)} & \Sexpr{prettyNum(ACU1$statistic)} & \Sexpr{prettyNum(ACU1$p.value)} \\
    Passive  & \Sexpr{prettyNum(mAC7.pr)} & \Sexpr{prettyNum(mAC7.po)} & \Sexpr{prettyNum(ACU7$statistic)} & \Sexpr{prettyNum(ACU7$p.value)} \\ \hline
\end{tabular}
\end{table}

\begin{table}
  \caption{Dryer}
  \begin{tabular}{c c c c c}
    Condition & Pre Rank Dev & Post Rank Dev & Statistic & p-value \\ \hline
    \$ only   & \Sexpr{prettyNum(mdry5.pr)} & \Sexpr{prettyNum(mdry5.po)} & \Sexpr{prettyNum(DRU5$statistic)} & \Sexpr{prettyNum(DRU5$p.value)} \\
    kWh only   & \Sexpr{prettyNum(mdry6.pr)} & \Sexpr{prettyNum(mdry6.po)} & \Sexpr{prettyNum(DRU6$statistic)} & \Sexpr{prettyNum(DRU6$p.value)} \\
    \$ and kWh & \Sexpr{prettyNum(mdry2.pr)} & \Sexpr{prettyNum(mdry2.po)} & \Sexpr{prettyNum(DRU2$statistic)} & \Sexpr{prettyNum(DRU2$p.value)} \\
    \$ app.   & \Sexpr{prettyNum(mdry3.pr)} & \Sexpr{prettyNum(mdry3.po)} & \Sexpr{prettyNum(DRU3$statistic)} & \Sexpr{prettyNum(DRU3$p.value)} \\
    kWh app.   & \Sexpr{prettyNum(mdry4.pr)} & \Sexpr{prettyNum(mdry4.po)} & \Sexpr{prettyNum(DRU4$statistic)} & \Sexpr{prettyNum(DRU4$p.value)} \\
    \$ and kWh app.   & \Sexpr{prettyNum(mdry1.pr)} & \Sexpr{prettyNum(mdry1.po)} & \Sexpr{prettyNum(DRU1$statistic)} & \Sexpr{prettyNum(DRU1$p.value)} \\
    Passive  & \Sexpr{prettyNum(mdry7.pr)} & \Sexpr{prettyNum(mdry7.po)} & \Sexpr{prettyNum(DRU7$statistic)} & \Sexpr{prettyNum(DRU7$p.value)} \\ \hline
\end{tabular}
\end{table}

\begin{table}
  \caption{Oven}
  \begin{tabular}{c c c c c}
    Condition & Pre Rank Dev & Post Rank Dev & Statistic & p-value \\ \hline
    \$ only   & \Sexpr{prettyNum(mov5.pr)} & \Sexpr{prettyNum(mov5.po)} & \Sexpr{prettyNum(OVU5$statistic)} & \Sexpr{prettyNum(OVU5$p.value)} \\
    kWh only   & \Sexpr{prettyNum(mov6.pr)} & \Sexpr{prettyNum(mov6.po)} & \Sexpr{prettyNum(OVU6$statistic)} & \Sexpr{prettyNum(OVU6$p.value)} \\
    \$ and kWh & \Sexpr{prettyNum(mov2.pr)} & \Sexpr{prettyNum(mov2.po)} & \Sexpr{prettyNum(OVU2$statistic)} & \Sexpr{prettyNum(OVU2$p.value)} \\
    \$ app.   & \Sexpr{prettyNum(mov3.pr)} & \Sexpr{prettyNum(mov3.po)} & \Sexpr{prettyNum(OVU3$statistic)} & \Sexpr{prettyNum(OVU3$p.value)} \\
    kWh app.   & \Sexpr{prettyNum(mov4.pr)} & \Sexpr{prettyNum(mov4.po)} & \Sexpr{prettyNum(OVU4$statistic)} & \Sexpr{prettyNum(OVU4$p.value)} \\
    \$ and kWh app.   & \Sexpr{prettyNum(mov1.pr)} & \Sexpr{prettyNum(mov1.po)} & \Sexpr{prettyNum(OVU1$statistic)} & \Sexpr{prettyNum(OVU1$p.value)} \\
    Passive  & \Sexpr{prettyNum(mov7.pr)} & \Sexpr{prettyNum(mov7.po)} & \Sexpr{prettyNum(OVU7$statistic)} & \Sexpr{prettyNum(OVU7$p.value)} \\ \hline
\end{tabular}
\end{table}

\begin{table}
  \caption{Microwave}
  \begin{tabular}{c c c c c}
    Condition & Pre Rank Dev & Post Rank Dev & Statistic & p-value \\ \hline
    \$ only   & \Sexpr{prettyNum(mmi5.pr)} & \Sexpr{prettyNum(mmi5.po)} & \Sexpr{prettyNum(MIU5$statistic)} & \Sexpr{prettyNum(MIU5$p.value)} \\
    kWh only   & \Sexpr{prettyNum(mmi6.pr)} & \Sexpr{prettyNum(mmi6.po)} & \Sexpr{prettyNum(MIU6$statistic)} & \Sexpr{prettyNum(MIU6$p.value)} \\
    \$ and kWh & \Sexpr{prettyNum(mmi2.pr)} & \Sexpr{prettyNum(mmi2.po)} & \Sexpr{prettyNum(MIU2$statistic)} & \Sexpr{prettyNum(MIU2$p.value)} \\
    \$ app.   & \Sexpr{prettyNum(mmi3.pr)} & \Sexpr{prettyNum(mmi3.po)} & \Sexpr{prettyNum(MIU3$statistic)} & \Sexpr{prettyNum(MIU3$p.value)} \\
    kWh app.   & \Sexpr{prettyNum(mmi4.pr)} & \Sexpr{prettyNum(mmi4.po)} & \Sexpr{prettyNum(MIU4$statistic)} & \Sexpr{prettyNum(MIU4$p.value)} \\
    \$ and kWh app.   & \Sexpr{prettyNum(mmi1.pr)} & \Sexpr{prettyNum(mmi1.po)} & \Sexpr{prettyNum(MIU1$statistic)} & \Sexpr{prettyNum(MIU1$p.value)} \\
    Passive  & \Sexpr{prettyNum(mmi7.pr)} & \Sexpr{prettyNum(mmi7.po)} & \Sexpr{prettyNum(MIU7$statistic)} & \Sexpr{prettyNum(MIU7$p.value)} \\ \hline
\end{tabular}
\end{table}


\begin{table}
  \caption{Heater}
  \begin{tabular}{c c c c c}
    Condition & Pre Rank Dev & Post Rank Dev & Statistic & p-value \\ \hline
    \$ only   & \Sexpr{prettyNum(mhe5.pr)} & \Sexpr{prettyNum(mhe5.po)} & \Sexpr{prettyNum(HEU5$statistic)} & \Sexpr{prettyNum(HEU5$p.value)} \\
    kWh only   & \Sexpr{prettyNum(mhe6.pr)} & \Sexpr{prettyNum(mhe6.po)} & \Sexpr{prettyNum(HEU6$statistic)} & \Sexpr{prettyNum(HEU6$p.value)} \\
    \$ and kWh & \Sexpr{prettyNum(mhe2.pr)} & \Sexpr{prettyNum(mhe2.po)} & \Sexpr{prettyNum(HEU2$statistic)} & \Sexpr{prettyNum(HEU2$p.value)} \\
    \$ app.   & \Sexpr{prettyNum(mhe3.pr)} & \Sexpr{prettyNum(mhe3.po)} & \Sexpr{prettyNum(HEU3$statistic)} & \Sexpr{prettyNum(HEU3$p.value)} \\
    kWh app.   & \Sexpr{prettyNum(mhe4.pr)} & \Sexpr{prettyNum(mhe4.po)} & \Sexpr{prettyNum(HEU4$statistic)} & \Sexpr{prettyNum(HEU4$p.value)} \\
    \$ and kWh app.   & \Sexpr{prettyNum(mhe1.pr)} & \Sexpr{prettyNum(mhe1.po)} & \Sexpr{prettyNum(HEU1$statistic)} & \Sexpr{prettyNum(HEU1$p.value)} \\
    Passive  & \Sexpr{prettyNum(mhe7.pr)} & \Sexpr{prettyNum(mhe7.po)} & \Sexpr{prettyNum(HEU7$statistic)} & \Sexpr{prettyNum(HEU7$p.value)} \\ \hline
\end{tabular}
\end{table}


\begin{table}
  \caption{Washer}
  \begin{tabular}{c c c c c}
    Condition & Pre Rank Dev & Post Rank Dev & Statistic & p-value \\ \hline
    \$ only   & \Sexpr{prettyNum(mwa5.pr)} & \Sexpr{prettyNum(mwa5.po)} & \Sexpr{prettyNum(WAU5$statistic)} & \Sexpr{prettyNum(WAU5$p.value)} \\
    kWh only   & \Sexpr{prettyNum(mwa6.pr)} & \Sexpr{prettyNum(mwa6.po)} & \Sexpr{prettyNum(WAU6$statistic)} & \Sexpr{prettyNum(WAU6$p.value)} \\
    \$ and kWh & \Sexpr{prettyNum(mwa2.pr)} & \Sexpr{prettyNum(mwa2.po)} & \Sexpr{prettyNum(WAU2$statistic)} & \Sexpr{prettyNum(WAU2$p.value)} \\
    \$ app.   & \Sexpr{prettyNum(mwa3.pr)} & \Sexpr{prettyNum(mwa3.po)} & \Sexpr{prettyNum(WAU3$statistic)} & \Sexpr{prettyNum(WAU3$p.value)} \\
    kWh app.   & \Sexpr{prettyNum(mwa4.pr)} & \Sexpr{prettyNum(mwa4.po)} & \Sexpr{prettyNum(WAU4$statistic)} & \Sexpr{prettyNum(WAU4$p.value)} \\
    \$ and kWh app.   & \Sexpr{prettyNum(mwa1.pr)} & \Sexpr{prettyNum(mwa1.po)} & \Sexpr{prettyNum(WAU1$statistic)} & \Sexpr{prettyNum(WAU1$p.value)} \\
    Passive  & \Sexpr{prettyNum(mwa7.pr)} & \Sexpr{prettyNum(mwa7.po)} & \Sexpr{prettyNum(WAU7$statistic)} & \Sexpr{prettyNum(WAU7$p.value)} \\ \hline
\end{tabular}
\end{table}


\begin{table}
  \caption{Freezer}
  \begin{tabular}{c c c c c}
    Condition & Pre Rank Dev & Post Rank Dev & Statistic & p-value \\ \hline
    \$ only   & \Sexpr{prettyNum(mfr5.pr)} & \Sexpr{prettyNum(mfr5.po)} & \Sexpr{prettyNum(FRU5$statistic)} & \Sexpr{prettyNum(FRU5$p.value)} \\
    kWh only   & \Sexpr{prettyNum(mfr6.pr)} & \Sexpr{prettyNum(mfr6.po)} & \Sexpr{prettyNum(FRU6$statistic)} & \Sexpr{prettyNum(FRU6$p.value)} \\
    \$ and kWh & \Sexpr{prettyNum(mfr2.pr)} & \Sexpr{prettyNum(mfr2.po)} & \Sexpr{prettyNum(FRU2$statistic)} & \Sexpr{prettyNum(FRU2$p.value)} \\
    \$ app.   & \Sexpr{prettyNum(mfr3.pr)} & \Sexpr{prettyNum(mfr3.po)} & \Sexpr{prettyNum(FRU3$statistic)} & \Sexpr{prettyNum(FRU3$p.value)} \\
    kWh app.   & \Sexpr{prettyNum(mfr4.pr)} & \Sexpr{prettyNum(mfr4.po)} & \Sexpr{prettyNum(FRU4$statistic)} & \Sexpr{prettyNum(FRU4$p.value)} \\
    \$ and kWh app.   & \Sexpr{prettyNum(mfr1.pr)} & \Sexpr{prettyNum(mfr1.po)} & \Sexpr{prettyNum(FRU1$statistic)} & \Sexpr{prettyNum(FRU1$p.value)} \\
    Passive  & \Sexpr{prettyNum(mfr7.pr)} & \Sexpr{prettyNum(mfr7.po)} & \Sexpr{prettyNum(FRU7$statistic)} & \Sexpr{prettyNum(FRU7$p.value)} \\ \hline
\end{tabular}
\end{table}

\begin{table}
  \caption{TV}
  \begin{tabular}{c c c c c}
    Condition & Pre Rank Dev & Post Rank Dev & Statistic & p-value \\ \hline
    \$ only   & \Sexpr{prettyNum(mTV5.pr)} & \Sexpr{prettyNum(mTV5.po)} & \Sexpr{prettyNum(TVU5$statistic)} & \Sexpr{prettyNum(TVU5$p.value)} \\
    kWh only   & \Sexpr{prettyNum(mTV6.pr)} & \Sexpr{prettyNum(mTV6.po)} & \Sexpr{prettyNum(TVU6$statistic)} & \Sexpr{prettyNum(TVU6$p.value)} \\
    \$ and kWh & \Sexpr{prettyNum(mTV2.pr)} & \Sexpr{prettyNum(mTV2.po)} & \Sexpr{prettyNum(TVU2$statistic)} & \Sexpr{prettyNum(TVU2$p.value)} \\
    \$ app.   & \Sexpr{prettyNum(mTV3.pr)} & \Sexpr{prettyNum(mTV3.po)} & \Sexpr{prettyNum(TVU3$statistic)} & \Sexpr{prettyNum(TVU3$p.value)} \\
    kWh app.   & \Sexpr{prettyNum(mTV4.pr)} & \Sexpr{prettyNum(mTV4.po)} & \Sexpr{prettyNum(TVU4$statistic)} & \Sexpr{prettyNum(TVU4$p.value)} \\
    \$ and kWh app.   & \Sexpr{prettyNum(mTV1.pr)} & \Sexpr{prettyNum(mTV1.po)} & \Sexpr{prettyNum(TVU1$statistic)} & \Sexpr{prettyNum(TVU1$p.value)} \\
    Passive  & \Sexpr{prettyNum(mTV7.pr)} & \Sexpr{prettyNum(mTV7.po)} & \Sexpr{prettyNum(TVU7$statistic)} & \Sexpr{prettyNum(TVU7$p.value)} \\ \hline
\end{tabular}
\end{table}

\begin{table}
  \caption{Fridge}
  \begin{tabular}{c c c c c}
    Condition & Pre Rank Dev & Post Rank Dev & Statistic & p-value \\ \hline
    \$ only   & \Sexpr{prettyNum(mFD5.pr)} & \Sexpr{prettyNum(mFD5.po)} & \Sexpr{prettyNum(FDU5$statistic)} & \Sexpr{prettyNum(FDU5$p.value)} \\
    kWh only   & \Sexpr{prettyNum(mFD6.pr)} & \Sexpr{prettyNum(mFD6.po)} & \Sexpr{prettyNum(FDU6$statistic)} & \Sexpr{prettyNum(FDU6$p.value)} \\
    \$ and kWh & \Sexpr{prettyNum(mFD2.pr)} & \Sexpr{prettyNum(mFD2.po)} & \Sexpr{prettyNum(FDU2$statistic)} & \Sexpr{prettyNum(FDU2$p.value)} \\
    \$ app.   & \Sexpr{prettyNum(mFD3.pr)} & \Sexpr{prettyNum(mFD3.po)} & \Sexpr{prettyNum(FDU3$statistic)} & \Sexpr{prettyNum(FDU3$p.value)} \\
    kWh app.   & \Sexpr{prettyNum(mFD4.pr)} & \Sexpr{prettyNum(mFD4.po)} & \Sexpr{prettyNum(FDU4$statistic)} & \Sexpr{prettyNum(FDU4$p.value)} \\
    \$ and kWh app.   & \Sexpr{prettyNum(mFD1.pr)} & \Sexpr{prettyNum(mFD1.po)} & \Sexpr{prettyNum(FDU1$statistic)} & \Sexpr{prettyNum(FDU1$p.value)} \\
    Passive  & \Sexpr{prettyNum(mFD7.pr)} & \Sexpr{prettyNum(mFD7.po)} & \Sexpr{prettyNum(FDU7$statistic)} & \Sexpr{prettyNum(FDU7$p.value)} \\ \hline
\end{tabular}
\end{table}

\begin{table}
  \caption{Indoor Lights}
  \begin{tabular}{c c c c c}
    Condition & Pre Rank Dev & Post Rank Dev & Statistic & p-value \\ \hline
    \$ only   & \Sexpr{prettyNum(mIN5.pr)} & \Sexpr{prettyNum(mIN5.po)} & \Sexpr{prettyNum(INU5$statistic)} & \Sexpr{prettyNum(INU5$p.value)} \\
    kWh only   & \Sexpr{prettyNum(mIN6.pr)} & \Sexpr{prettyNum(mIN6.po)} & \Sexpr{prettyNum(INU6$statistic)} & \Sexpr{prettyNum(INU6$p.value)} \\
    \$ and kWh & \Sexpr{prettyNum(mIN2.pr)} & \Sexpr{prettyNum(mIN2.po)} & \Sexpr{prettyNum(INU2$statistic)} & \Sexpr{prettyNum(INU2$p.value)} \\
    \$ app.   & \Sexpr{prettyNum(mIN3.pr)} & \Sexpr{prettyNum(mIN3.po)} & \Sexpr{prettyNum(INU3$statistic)} & \Sexpr{prettyNum(INU3$p.value)} \\
    kWh app.   & \Sexpr{prettyNum(mIN4.pr)} & \Sexpr{prettyNum(mIN4.po)} & \Sexpr{prettyNum(INU4$statistic)} & \Sexpr{prettyNum(INU4$p.value)} \\
    \$ and kWh app.   & \Sexpr{prettyNum(mIN1.pr)} & \Sexpr{prettyNum(mIN1.po)} & \Sexpr{prettyNum(INU1$statistic)} & \Sexpr{prettyNum(INU1$p.value)} \\
    Passive  & \Sexpr{prettyNum(mIN7.pr)} & \Sexpr{prettyNum(mIN7.po)} & \Sexpr{prettyNum(INU7$statistic)} & \Sexpr{prettyNum(INU7$p.value)} \\ \hline
\end{tabular}
\end{table}

Unexpectedly, the accuracy of ranking of two appliances were actually harmed by the treatment.  Oven was \emph{harmed} by total \$ only, but no other conditions, $V=$ \Sexpr{prettyNum(OVU5$statistic)}, $p=$ \Sexpr{prettyNum(OVU5$p.value)}.  Heater was \emph{harmed} by total kWh only $V=$ \Sexpr{prettyNum(HEU6$statistic)}, $p=$ \Sexpr{prettyNum(HEU6$p.value)}.

\clearpage
\section{Detailed cost rank analyses}
\label{app:simcost}

As can be seen in Table~\ref{tab:costrank}, for the oven, participants were harmed by total kWh and \$ $V=$ \Sexpr{prettyNum(OVC2$statistic)}, $p=$ \Sexpr{prettyNum(OVC2$p.value)}, harmed by kWh by appliance $V=$ \Sexpr{prettyNum(OVC4$statistic)}, $p=$ \Sexpr{prettyNum(OVC4$p.value)}, but helped by passive learning $V=$ \Sexpr{prettyNum(OVC7$statistic)}, $p=$ \Sexpr{prettyNum(OVC7$p.value)}.  Heater benefitted in \$ and kWh by appliance $V=$ \Sexpr{prettyNum(HEC1$statistic)}, $p=$ \Sexpr{prettyNum(HEC1$p.value)}, and passive learning $V=$ \Sexpr{prettyNum(HEC7$statistic)}, $p=$ \Sexpr{prettyNum(HEC7$p.value)}.  For the washing machine benefitted from kWh by appliance $V=$ \Sexpr{prettyNum(WAC4$statistic)}, $p=$ \Sexpr{prettyNum(WAC4$p.value)}, total kWh only $V=$ \Sexpr{prettyNum(WAC6$statistic)}, $p=$ \Sexpr{prettyNum(WAC6$p.value)}, and passive learning $V=$ \Sexpr{prettyNum(WAC7$statistic)}, $p=$ \Sexpr{prettyNum(WAC7$p.value)}.  The freezer only benefitted from passive learning $V=$ \Sexpr{prettyNum(FRC7$statistic)}, $p=$ \Sexpr{prettyNum(FRC7$p.value)}.  TV benefitted from kWh and \$ by appliance $V=$ \Sexpr{prettyNum(TVC1$statistic)}, $p=$ \Sexpr{prettyNum(TVC1$p.value)}, and \$ only by appliance $V=$ \Sexpr{prettyNum(TVC3$statistic)}, $p=$ \Sexpr{prettyNum(TVC3$p.value)}.  Indoor lights were harmed by \$ and kWh by appliance $V=$ \Sexpr{prettyNum(INC1$statistic)}, $p=$ \Sexpr{prettyNum(INC1$p.value)}, but helped by passive learning $V=$ \Sexpr{prettyNum(INC7$statistic)}, $p=$ \Sexpr{prettyNum(INC7$p.value)}.  There were no significant benefits for cost for AC, dryer, microwave, refrigerator.

 \begin{table}[h]
  \centering
\scalebox{0.8}{
  \begin{tabular}{c c c c c c c c c}
 Appliance    
 & \$, kWh, AS 
 & \$, kWh 
 & \$, AS 
 & kWh, AS 
 & \$ 
 & kWh 
 & Passive & Total \\ \hline
AC            &  &  &  &  &  &  &  &   \\
Dryer         &  &  &  &  &  &  &  &   \\
Microwave     &  &  &  &  &  &  &  &   \\
Oven          &  &- &  &- &  &  &+ &   \\
Water Heater  &+ &  &  &  &  &  &+ &   \\
TV            &+ &  &  &  &+ &  &  &   \\ 
Fridge        &  &  &  &  &  &  &  &   \\ 
Indoor Lights &- &  &  &  &  &  &+ &   \\ 
Washer        &  &  &  &+ &  &+ &+ &   \\ 
Freezer       &  &  &  &  &  &  &+ &   \\ \hline
Total         &2 (-1) & (-1) & & 0 & 1 (-1) & 1 & 5 &  ? \\ \hline  
  \end{tabular}}
  \caption{Helped (+) or harmed (-) rankings of energy cost for each appliance. AS is appliance specific feedback.}
\label{tab:costrank}
 \end{table}
 
 Thus, in terms of learning the cost each appliance contributes to the monthly bill, only passive learning was effective, with some indication that kWh and \$ appliance specific information also help.

 
%\bibliographystyle{apalike}
\bibliographystyle{unsrt}
\bibliography{/home/alex/Dropbox/masterbib}
\end{document}

In general, the expression of preferences in self-report is only valid if those preferences are associated with other important criteria, like behavior (Roberts and Baker, 2003).  Innovative methods of collecting data on electricity behaviors and the development of valid self-reports is lacking (Steg and Vlek, 2009).  More valid measures can be obtained using psychometric modeling, such as the General Structural Model (DeVellis, 2011), quizzes, (Roberts and Baker, 2003), and Item-Response Theory (DeMars, 2010).  We begin the development of the Standardized Electricity Knowledge Test (SEKT). 


IHD sim manuscript (initial survey and IHD sim experiment 
Don't seem to be using the simpe 'N manipulations' strategy
People don't know what a hot water heater is
Might be able to deduce the rank of appliances; using reasoning (
Confidence measures of rankings
Change in confidence before and after simulation in their rankings
Problem solving/reasoning task to separate high reasoners (or high need for cognition); IQ task
Talk to group about resolving Q824
Goal/Game: points awarded for accuracy in question answering: (oven used X kwh; how long was it run for?)
Make the task so they have to run all appliances for X hours, not X hours individually for each appliance.
IHD sim
construct validity
measurement
acceptance sampling; 10\% error rate

Ours is very close to Karjalainen, 2011 prototype 6.

Methods: binary choice (similarity scaling); likert rating; evaluation; attitudes; etc.
1) Passive vs. active learning: need a condition in the IHD sim where we just give them the answers
a. Learning from observations and interventions (Steyvers, et al, 2003; Solway and Botvinick, 2012; Holyoak and Cheng, 2011; Goodman et al, 2011; Rottman and Keil, 2012)
2) Detail the judgments we are asking them to make.  How do we think they are making these judgments (heuristics; Byesian; etc.?)
3) Try to differentiate the mechanisms

Communication

I’ve modified the survey so most questions/instructions are within tolerable limit of a flesh-kincaid reading level of 6th grade.  In the retrospective probes section, I’ve pinpointed specific questions/instructions that need to be carefully examined during the cognitive interviews

Paradigm and Experiment

We must keep in mind both what we are manipulating and what is remaining constant in the experiment.  We are manipulating three factors: 1) kWh feedback, 2) \$ feedback, and 3) appliance-specific info.  Evaluating our experimental intervention requires questioning whether these three manipulations do what they intend to do perfectly, or do they have effects only in a noisy or stochastic manner.  If the latter is the case, can we formulate a method of improving the effectiveness of the intervention?

We must also consider what does not change between participants.  All are exposed to repeated measures of the same questions, creating the possibility for testing effects.  None are exposed to repeated trials of the IHD, meaning they might have lacked a ‘warm-up’ period.

Internal Validity

I’ve highlighted the riskiest threats to the internal validity of the study.  The following are threats to the internal validity of the study.   If we can figure out ways of minimizing them, that would be great
1) Condition blinding: The participants know what the treatment is.
2) Hypothesis blinding:  Participants may be able to figure out what the hypothesis is.
3) Incomplete outcome data: How can we minimize dropout?
4) Research expectancies:  Have we designed the study in such a way as to lead participants to the answers we want?
5) Treatment diffusion:  Is it possible for people to complete the experiment more than once?
6) Testing:  Since measurements are repeated, we can have testing effects, such as an order effect.
7) Instability:  Are our instruments stable and reliable over time?
8) Maturation:  Is the task boring, thus creating poorer responses to the post-test?

Measurement
What are the measurement requirements of our questions.  Many of them are nominal; is that the right way to do it rather than ordinal?  Do we have any items that require an interval scale?  If so, how can we establish that the scale meets the conditions required for an interval scale?

External Valdity
There are serious threats to the external validity of this experiment.  We are drawing a convenience sample from a population different from the one we are interested in.  Is it possible to do random sampling of Pepco customers based on email?  This would drastically increase the external validity.  We must also consider the differences between the outcome measures we are considering here and how they will be measured in the field study.  Obviously we cannot measure peak/overall reduction in this experiment.  However, we should consider how we will measure learning and change in knowledge states in the field study and try to bring out measures in this experiment as close as possible to those we will use in the real world. 

Construct Validity
I’ve identified the following constructs and their associated questions.  We must think critically about the validity of the constructs and whether the specific questions (operations) capture the construct in the best way.  We should also consider at least the convergent and discriminant validity of the constructs.

The constructs I’ve identified are the following
1) Expected benefit (EB) (Q352; Q363; Q410; Q359; Q874)
2) Perceived learning (PL) (Q357)
3) Numeracy (Q287)
4) Knowledge of appliance usage (Q253/Q862; Q318/Q863; Q420/Q864; Q421/Q865; Q422/Q866; Q423/Q867)
5) Knowledge of peak time (Q319/Q868; Q827/Q869)
6) General knowledge of electricity (Q321/Q870; Q828/Q871)
7) Knowledge of electricity price (Q257/Q872)
8) Knowledge of efficiency behaviors (Q824; Q296; Q873)
I’ve modified the survey so most questions/instructions are within tolerable limit of a flesh-kincaid reading level of 6th grade.  In the retrospective probes section, I’ve pinpointed specific questions/instructions that need to be carefully examined during the cognitive interviews

Methods
Results
Constructs: dimensionality; internal consistency; item-whole correlations; structural equations
Write up hypothetical results
http://en.wikipedia.org/wiki/Spearman%E2%80%93Brown_prediction_formula
“In parallel forms reliability you first have to create two parallel forms. One way to accomplish this is to create a large set of questions that address the same construct and then randomly divide the questions into two sets. You administer both instruments to the same sample of people. The correlation between the two parallel forms is the estimate of reliability. One major problem with this approach is that you have to be able to generate lots of items that reflect the same construct. This is often no easy feat. Furthermore, this approach makes the assumption that the randomly divided halves are parallel or equivalent. Even by chance this will sometimes not be the case. The parallel forms approach is very similar to the split-half reliability described below. The major difference is that parallel forms are constructed so that the two forms can be used independent of each other and considered equivalent measures. For instance, we might be concerned about a testing threat to internal validity. If we use Form A for the pretest and Form B for the posttest, we minimize that problem. it would even be better if we randomly assign individuals to receive Form A or B on the pretest and then switch them on the posttest. With split-half reliability we have an instrument that we wish to use as a single measurement instrument and only develop randomly split halves for purposes of estimating reliability.”

Discussion

Study 4: Have Pepco Customers use the simulated IHD online & record their consumption
Now look at top and bottom special features and see if preferences match performance.
Highest scorer wins a prize (e.g., \$50)
was thinking soemthng like this
if we have sim/goal1->question 1; sim/goal2->question2...
and after each quiz they get a score that tells them how close they were AND how they score compared to everyone else
and they know at the end they will be told a total score and how they did compared to other people overall
Jack: ah ic that makes sense
Alex: then it could be a fun game
Jack: when you say how close they were
Alex: even one that yo could just put up on a website
Jack: what were thinking of
Alex: just like
Jack: like \% off from the solution?
Alex: the person spent \$10 on washing machine
how many loads did she do?
Jack: ohh
Alex: then the ccorrect answer is X
score them by the absolute difference
then they get like 6 scores
and they can see how well they did overall and on each six subscore
compared to other people
Jack: okay that sounds great
Alex: nd
Jack: it would use social comparison to make them motivated to do well
Alex: if we can randomly generate parameters
they could do it multiple times
and just put it on the web
nd let anyone take it
like fold-it or luis von ahn's human computation stuff
Jack: yeah that would be awesome
Alex: maybe throw that bye ashley and see if she can do something like that
Jack: how would we get it out to people?
okay, i'm going to email her today and ask her for a meeting
Alex: we would just make a website or put it on mturk

Email pepco customers with link to simulated IHD or placebo.  Different treatment groups have it mailed to them different # of times with different info/formats.  Have them use it several times. Measure consumption; collect data to do cognitive modeling and knowledge/attitude change; measure long-term effects

Field Study: have them predict/choose and then send them different email links with SIM learning  and measure consumption change.

To evaluate the relationship between preference and effectiveness, we highlight the performance-preference gap, where the customer’s expressed preference may not be best for meeting their desired goals (reducing energy consumption).  We measure both performance and subjective (preference) metrics. Performance measures include: success, time, errors, etc. Subjective measures include: user's self-reported satisfaction and comfort ratings.  People's performance and preference do not always match. Often users will perform poorly but their subjective ratings are very high. Conversely, they may perform well but subjective ratings are very low. (Experiment: have them choose what information they want to see and then randomly assigned of either their choice/nudger choice/random).

To evaluate the relationship between expectations and effectiveness, , we want to know accuracy and calibration of these three approaches in predicting treatment effects.  Norton (2008) found that participants perceived about a 5\% overall reduction in use, an overestimate compared to the real average reduction of about 3\%.  Similarly, Frank (2008) found a 6.7\% overall reduction in use whereas participants perceived an average of 9\% overall reduction.  For the segmentation, there may be groups of individuals who are likely to be able to develop the habitual action pattems necessary for these innovations” and “psychological elements that can be included in the design of these "action-requiring" conservation innovations that induce all people to use them successfully” (Seligman et al, 1979).

3.1 Procedure of creating IHD SIM
3.2 Participants
3.3 Results

Conclusions
Discussion
Next steps should be doing this kind of testing with multiple variables, then the creation of a real device and field-testing. This is an important first step in technology development, especially when technologies are likely to be implemented on a mass scale. Allows researchers and product developers to do systematic rigorous testing before providing a product at customer’s expense. 


Important theories exist, ranging from consumer choice, procedural and bounded rationality, to diffusion of innovation theories (Lopes et al, 2012).  Habits, consciousness, self-efficacy and feelings of control, and other motivations are very important and frequently mentioned (Fischer, 2008).  The heuristic model of environmentally relevant behavior involves norm activation (consciousness of a problem, consciousness of relevant behaviors, and consciousness of relevant possibilities), motivation (personal environmental norms, social norms, other motives such as satisficing), and evaluation (weighing all these inputs to determine actions) (Fischer, 2008). Contexts, habits, individual motivations, perceived costs and benefits, moral and social obligations, may be the most important (Lopes et al., 2012).v  Structural change, involving the rewards and benefits of the environment are also helpful in addition to informational change.


Research streams: cognitive psychology, social psychology, BDR, behavioral economics, ergonomics, human factors, industrial engineering, design, HCI.
Handbook of human factors and ergonomics (Salvendy, 2012)
Handbook of industrial engineering (Salvendy, 2007)
Handbook of behaviorl economics (Camerer et al, 2003?)
Design (Norman, 1988; Krug, 2007; Johnson, 2010)
HCI (Fu and Pirolli, 2007???)
Declarative and procedural knowledge (Anderson, 1983, 2007; Norman, 1988)

People can succeed in precise tasks because Information is in the world; great precision is not required; natural constraints are presnt; cultural constraints are present (Norman, 1988)

People glance, scan, and cli\usepackage{apacite}ck the first thing that seems interesting or seems like what we are looking for (Krug, 2007).  Skimming/scanning involves looking for matching phrases/words rather than reading (kind of like CTRL+F).  Satiscifing means we choose the first reasonable option tht we come across, rather than finding an exact match (Simon, XXXX; Klein, Sources of Power, 1999).  Finally, people muddle through haphazardly rather than exerting conscious effort to understand (Krug, 2007).  Muddling through means trial and error rather than reading the instructions. As long as we can get through it, that is enough; we don't need to understand. Effectively muddling through and 'getting it' can build a feeling of competence and maybe self-efficacy.

In-home displays seem to have an effect.  But why are they effective, and can they be designed to be more effective? 

\section{Why are IHDs effective?}

There are a number of ways in-home displays reduced use.  They can help people learn, form habits, and internalize energy-saving behaviors \citep{neenan2009residential,van1983patterns}.  People may become motivated to outperform oneself, others, or meet consumption targets through interacting with the display \cite{wood2007energy}.  

In this section, we discuss ten mechanisms through which IHDs can lead to energy reduction: 1) habits, 2) self-efficacy, 3) goals, 4) mapping Behaviors to consumption, 5) energy efficient appliance purchases, 6) control state maintenance, 7) encouraging sustainable behaviors, 8) awareness/consciousness, 9) play/games, 10) net-benefit calculations.  These mechansims are in addition to important background factors, such as incentive structures, individual values, attitudes, and norms \cite{ajzen1980understanding}, and socioeconomic factors \cite{lopes2012energy}, as well as simultaneous interventions that induce cognitive dissonance, goal setting, or social modeling \cite{osbaldiston2012environmental}.

\subsection{Habits}

Much electricity use behavior is habitual.  Habits are automated behaviors that reduce the time and mental effort required to act.  Every day people turn the lights on and off at the same times, use the dishwasher after dinner, do clothes during a specific time, and shower in the morning or night.  As a result, the variability in use of lighting during weekdays is quite small, but within-day variations are larger, with high use in the evenings, possibly increasing monotonically during the day depending on who is in the household (i.e., children or older people home all day, middle aged people going to work) \cite{bladh2008towards}.  As opposed to consciously maintained goals of completing actions, people engage in opportunistic acts, where they only complete an action if the opportunity arises, not engaging in any special effort to complete it otherwise \cite{norman2002design}.

However, habits can result in a rut where simple, easier solutions are not discovered or even sought out.  Thus, changing habits is important but difficult (Gifford, 2011).  Habits may need to be distrupted and replaced for energy savings to be possible \cite{fischer2008feedback}.  Habit strength and automatic, procedural goals are habits using response-frequency measures \cite{aarts2000habits,steg2009encouraging}.  

\subsection{Self-Efficacy}

Self-efficacy proposes that meta-cognitive evaluations and self-regulation matter.  Getting things done increases our perception that we are able to affect and control the environment to realize our intentions.  This gives people ``self-satisfaction and a sense of pride and self-worth'' and as a consequence they ``refrain from behaving in ways that give rise to self-dissatisfaction, self-devaluation, and self-censure.'' \cite{bandura2001social}.  A well designed display that is effective and easy to use can increase motivation by increasing feelings of self-efficacy and competence \cite{thogersen2010electricity,bandura2001social}.  Devices that reduce the gulf between intentions and allowable actions, as well as make it easy to understand what the current state of consumption is, are likely to be very effective by enhancing self-efficacy \cite{norman2002design}.

Learned helplessness, on the other hand, is a pervasive feeling that one cannot control one's environment (Abramson, Seligman, and Teasdale, 1978; Lazarus and Folkman, 1984; Weiner, 1985; Dweck and Legett, 1988; Deci and Ryan, 2000).  A poorly designed device will make people feel that they are helpless in understanding and manipualting their electricity use, where they feel that.  As a result people will stop interacting with the device, because,``if you fail at something, you think it is your fault.  Therefore you think you cant do that task.  As a result, next time you have to do the task, you believe you cant so you don't even try.  The result is that you cant, just as you thought. You're trapped in a self-fulfilling prophecy'' \cite{norman2002design}.  A complex, confusing, and uninformative display design is worse than no display at all, as bad design reinforces learned helplessenss and undermines self-efficacy.  If saving energy is difficult, is full of obstacles, or information seems hard to comprehend, people with high self-efficacy are likely to persevere and succeed, resulting in energy savings, wereas those who believe they cannot do this, or subscribe to a fatalist perspective (Gifford, 2011), are unlikely to see benefits from the IHD.  If one feels it is too difficult to reduce wasted electricity through intentional action, then electricity may be “squandered” \cite{thogersen2010electricity}.  

\subsection{Goals}

There is very strong evidence on the effectiveness of goals.  Studies involving a large number of participants, using a variety of tasks, conducted in many different countries “show that setting specific and difficult goals results in better performance than instructing participants to do their best (irrespective of whether the goals are self-set or set by an external source)” (Wood and Newborough, 2007).  Goals can be set as hedonic goal-frames (e.g., enjoying leisure), gain goal-frames (e.g., save money), and normative goal-frames (e.g., do what the neighbors would approve of) (Steg and Vlek, 2009).  Goals should not be too small, too large, and need to be customized to the potential savings of the household, and should be accompanied with advice or tips for meeting the goal, and feedback on whether they are meeting it (Karjalainen, 2011).  The source of the goal and the timeframe used to achieve the goal may matter (Consolvo, 2009).  When failing ot meet a goal, people adjust their goal downward to meet their performance, and this decrement is larger than the increase in performance goals when meeting their goals (Illies, 2005).  The difficulty of self-set goals and commitment to achieving them is related to conscientiousness (Klein, 2006).  Goal setting is also likely related to perceptions about whether one needs to meet a goal or not,; that is whether one see's one's behavior as wasteful or inefficient.  If goals are challenging and no goal progress is evident, then participants may become discouraged and withdraw from task effort, and commitment in the face of failure to meet one's goal is related to self-efficacy (West, 2005).  Thus, for people low in self-efficacy it might be necessary only to give people goal feedback when they are performing well to keep them motivated.  Feedback that one is approaching a specific goal successfully can motivate attaining that goal.  Locke (2002) argues that the more difficult the goal, the greater the effort toward that goal, and that specific goals do not improve performance but reduce variability in performance.  Expectations about whether one can achieve one's goals matters for performance, where expecting that one cannot achieve a goal harms performance, but the difficulty of the goal matters more.  Self-efficacy is likely to matter more for goals that are self-set than goals that are set by another person.  Manageable short term goals are more effective than when aggregated together into a single long-term goal (McMillian and Sparkes, XXXX?).  Hard goals result in higher performance than "do your best" or vague goals is related to the ambiguity inherent in vague goals (Locke and Latham, 1990). This ambiguity allows individuals to justify to themselves that they have tried hard enough at a point that falls lower than the performance level of someone who is trying for a specific and challenging goal.

Nothwehr, 2006; This study examines whether changes in goal setting frequency predict changes in use of behavioral strategies over time, controlling for baseline strategy use, demographics and whether a person was trying to lose weight. What will happen if we have participants set reduction goals more frequently than just once a month?  Would people adjust their goals relative to their confidence in attaining the set goal? Would it lead people to reduce less than they can actually reduce because they're undercutting their reduction goals in fears that they won't attain them?  What happens when the reduction goals they set are very miniscule (i.e. reduce by 0.5\% over the course of a day/month) Would they actually hit them because they're attainable? Or would those reduction goals become so small that they don't want to put in the effort to reduce?  Goal setting frequency was found to be strongly and positively associated with use of the strategies measured, both at baseline and over-time.  Goal setting specifically related to diet or physical activity was, in most cases, more strongly associated with the corresponding strategies than goal setting related to body weight.  Self-monitoring appears to be quite strongly associated with goal setting frequency.  Results suggest that setting more specific goals for diet or physical activity is generally more strongly associated with strategy use than setting weight-related goals.

Atkinson (1958) – task difficulty, measured as a probability of task success, was related to performance in a curvilinear, inverse function. The highest level of effort occurred when the task was moderately difficult, and the lowest levels occurred when the task was either very easy or very hard. (What kind of tasks were people asked to perform?) Goal difficulty effect sizes (d) in meta-analyses ranged from 0.52 to 0.82 (Locke & Latham, 1990). Performance leveled off or decreased only when the limits of ability were reached or when commitment to a highly difficult goal lapsed (Erez & Zidon, 1984). Goal specificity reduces variation in performance.  How is goal difficulty measured for energy saving tasks? Do people who sign up for energy conservation programs imagine difficulty level to be easy, but when put in a field setting consider goals to be difficult? Does difficulty level of these goals increase or decrease in time? 

Social-cognitive theory: self-efficacy (task-specific confidence) is measured by getting efficacy ratings across a whole range of possible performance outcomes rather than from a single outcome. When goals are self-set, people with high self-efficacy set higher goals than do people with lower self-efficacy. They are also more committed to assigned goals, find and use better task strategies to attain the goals, and respond more positively to negative feedback than do people with low self-efficacy (Locke & Latham, 1990; Seijts & B. W. Latham, 2011). 

When people are trained in proper strategies, those given specific high-performance goals are more likely to use those strategies than people given other types of goals; hence, their performance improves (Earley & Perry, 1987).

Goal-setting theory appears to contradict Vroom's (1964) valence-instrumentality-expectancy theory. It states that the force to act is a multiplicative combination of valence (anticipated satisfaction), instrumentality (the belief that performance will lead to rewards), and expectancy  (the belief that effort will lead to performance needed to attain the rewards).

Directive function: goals direct attention and effort toward goal-relevant activities and away from goal-irrelevant activities. Rothkopf and Billington (1979) found students with specific learning goals paid attention  to and learned goal-relevant prose passages better than goal-irrelevant passages. 

Locke and Bryan (1969) people who were given feedback about multiple aspects of their performance on an automobile-driving task improved their performance on the dimensions for which they had goals but not on other dimensions.

Energizing function: high goals lead to greater effort than low goals. This has been shown with tasts that directly entail physical effort, entail repeated performance of simple cognitive tasks, such as addition, include measurements of subjective effort and include physiological indicators of effort.  

Goals affect persistence: when participants are allowed to control the time they spend on a task, hard goals prolong effort (LaPorte & Nath, 1976). There is often a trade-off in work between time and intensity of effort. Faced with a difficult goal, it is possible to work faster and more intensely for a long period. Tight deadlines lead to a more rapid work pace than loose deadlines in the laboratory (Bryan & Locke, 1967b) as well as in the field (Latham & Locke, 1975). 

Goals affect action indirectly by leading to the arousal, discovery, and/or use of task-relevant knowledge and strategies  When confronted with task goals, people automatically use the knowledge and skills they have already acquired that are relevant to goal attainment.  If the path to the goal is not a matter of using automatized skills, people draw from a repertoire of skills that they have used previously in related contexts, and they apply them to the present situation. 
If the task for which a goal is assigned is new to people, they will engage in deliberate planning to develop strategies that will enable them to attain their goals (Smith, Locke, & Barry, 1990).

People with high self-efficacy are more likely than those with low self-efficacy to develop effective task strategies (Latham, Winters, & Locke, 1994; Wood & Bandura, 1989). There may be a time lag between assignment of the goal and the effects of the goal on performance, as people search for appropriate strategies (Smith et al., 1990). 

When people are confronted with a task that is complex for them, urging them to do their best sometimes leads to better strategies (Earley, Connolly, & Ekegren, 1989) than setting a specific difficult performance goal. Setting specific challenging learning goals, such as to discover a certain number of different strategies to master the task (Seijts & G. P. Latham, 2001; Winters & Latham, 1996). Does this apply to learning goals?

Covington, 2000

\subsection{Mapping Behaviors to Consumption}

When customers learn what actions they can take, and how each action affects their electricity use, they map behaviors to consumption.  Rather than simply increasing awareness by ``drawing attention to the cost of energy'' \cite{mcclelland1979energy}, the IHD ``teach[es] residents what activities consume the most energy'' .  People interacting with a device initially did not know what used the most electricity, ``P1 --- realized the A/C uses a lot more power than he initially suspected; P3 was surprised that the TV didn't ---have a whole lot of effect; and, P5 did not anticipate that the clothes dryer would have such significant impact on consumption''\cite{yun2009investigating}.

There is conflicting evidence, however.  People have difficutly figuring out what needs to be done, even if they know what uses the most energy.  That is they ``do not quite know what they can do to reduce their electricity consumption'' \cite{eiden2009investigation}.  If this is the case, then energy-saving tips would have to accompany the IHD for it to be effective.

\subsection{Energy Efficient Appliance Purchases}

In-home displays may work indirectly by encouraging energy efficient appliance purchases, but otherwise have no effect on curtailment behaviors.  By learning about total consumption, cost, and appliance-specific use, households may decide that the best course of action is to purchase an energy efficient appliance.  This leads to a ``one-time decision to initiate the retrofit'' \cite{seligman1978behavioral}.  \citeA{dobson1992conservation} found that users of the RECS knew more about what appliances consume energy and that this caused them to purchase more efficient appliances.  \citeA{yun2009investigating} found that two households replaced high-power consuming devices with more energy efficient devices.

\subsection{Control State Maintenance}

Because electricity is consumed implicitly in a variety of contexts and modes, there is no ``concise cognitive frame'' of electricity consumption that makes sense and is usable to people \cite{fischer2008feedback}.  In-home displays can help give context to the user's electricity consumption. The IHD helps consumers maintain an optimal consumption state and avoid waste \cite{seligman1978behavioral}.  Control states focus peoples' attention to specific actions and exactly when they are appropriate.  The IHD may simplify the energy consumption tasks by telling customers exactly what to do and when.  \citet{yun2009investigating} observed of his participants: ``P2 independently made the decision to keep his ECD between 2 and 3 lights.  When the display was higher, he reported setting out to investigate, turning off devices along the way.  P5 reduced her household's average daily consumption by more than 50\% by pursuing the goal of never having the ECD blink.''  It may make a difference whether the control states or goals are self-imposed or encouraged externally \cite{yun2009investigating}.

Carver and Scheier, 1981 action theory (e.g., Frese & Zapf, 1994),

Frese, M., & Zapf, D. (1994). Action as the core of work psychology: A German approach. In H. C. Triandis, M. D. Dunnette, & L. M. Hough (Eds.), Handbook of industrial and organizational psychology (2nd ed., Vol. 4, pp. 271-340). Palo Alto, CA: Consulting Psychologists Press.
action identification theory (Vallacher & Wegner, 1987)
Vallacher, R. R., & Wegner, D. M. (1987). What do people think they're doing? Action identification and human behavior. Psychological Review, 94, 3-15.
variant of learned helplessness theory (e.g., Mikulincer, 1994),

\subsection{Encouraging Sustainable Behaviors}

The in-home display may encourage sustainable behaviors by helping people engage in behaviors that they would like to commit to but have thus far failed to implement (\cite{yun2009investigating} Chetty et al, 2008; getting to green). [6] McCalley, L. T. From motivation and cognition theories to everyday applications and back again: the case of product-integrated information and feedback. Energy Policy 34, no. 2 (January 2006): 129-137.  \citeA{paetz2011shifting} found that although habits did not change, attitudes toward electricity use did, making people feel guilty about wasting electricity, ``This load curve has changed my attitudes. At least I know how much power the coffee machine needs. This morning for example I turned it off, because I knew my roommate was still asleep and I thought it doesn't have to run for another hour without need. (T1 interview)'' \cite{paetz2011shifting}.

\subsection{Awareness/Consciousness}

If people do not know to turn off a light, or that a water heater has a temperature setting, then they cannot change their behavior to conserve energy (Gifford, 2011).  In-Home Displays are likely to increase awareness of appliances and energy-saving behaviors, as people realize that they are consuming ``an invisible product that is often ignored'' \cite{schembri2008influence}. 

This is also a plausible explanation of the data.  Awareness, and not the content of the display itself, explains why more sophisticated feedback (e.g., real-time, graphical displays) hasn't been associated with larger effect sizes, ``It seems that it is the presence of the information itself --- not its presentation in a more salient, graphical format - that is causing the behavior change'' \cite{allen2006effects}.  \citeA{yun2009investigating} found that those who had low or moderate awareness of energy consumption reduced their energy consumption more than those who had high awareness initially, suggesting their awareness was raised more by the display.  \citeA{hutton1986effects} found largest effects of the ECI above and beyond education alone from participants who knew the least in California as compared to two Canadian cities.  Norton et al (2008) found that 48\% of participants reported increased awareness of energy efficiency actions somewhat, 27\% significantly, and 23\% remained the same.  

However, consciousness of problematic electricity use, behaviors that contribute to them, and ways of fixing this is necessary, but not sufficient for behavioral change \cite{fischer2008feedback}.  

\subsection{Play/Games}

Interacting with the IHD and saving energy can be fun.  Some households naturally create energy saving games, such as trying to find ``how low can you go'' in electricity consumption \cite{yun2009investigating}.  Another example of game play comes from participants living in a smart energy home.  They blogged their daily activities in the smart home, engaging in play behavior: ``For half an hour I have turned on as many appliances as possible, even my hair curler. I was impressed by 7000 Watts and no shortage ;-), but shocked that the hoover needed 4000 Watts power. It's like a game.'' (T2 blog) \cite{paetz2011shifting}.  Thus, in-home displays may be effective to the extent that they promote an environment for fun or ability to play with one's electricity through manipulation and feedback.  The device can also include pre-programmed games, as the ECI did \cite{hutton1986effects}.  

Garris and Ahlers, 2002

\subsection{Net-Benefit Calculations}

The perceived costs and benefits of curtailing electricity use will likely determine whether those behaviors are undertaken \cite{mckenzie2011fostering}. People intuitively perform net-benefit calculation calculations of behaviors.  When the effort and financial costs required to take energy conservation acts outweighs the perceived benefits, people won't engage in the behavior.  For example, in interviews, residential customers found that they could quickly identify and provide a ``comprehensive list of energy saving measures (e.g., cavity wall insulation) and behavioural changes (e.g., turning your thermostat down by 1 degree C)'', but didn't do it because net-benefit was perceived to be negative \cite{roberts2004consumer}.  In-home displays can correct inaccurate perceptions of net cost and benefit, thus promoting conservation behaviors \cite{steg2009encouraging}.

\section{How can IHDs be designed to be more effective?}

There is a complete lack of information on what aspects of in-home displays help customers learn and save energy, although there is general agreement that the feedback should be attention-grabbing, clear, appealing, and specific \cite{karjalainen2011consumer}.  However, many studies have looked at \emph{preferences} for information.  

Because there is no direct empirical or experimental evidence on how people learn from IHDs, we look more generally.  In this section we look at taxonomies, design principles (both general and specific), and heterogenous agents.


\subsection{General Design Principles}

Given the vast range of possibilities that the taxonomy allows, it is helpful to narrow the range using general principles of design.  In general, attention should be paid to selecting the right behaviors to target, what causes those behaviors, and measuring the effectiveness of interventions based on whether the intervention did what it was supposed to do \cite{steg2009encouraging}. 

Some guidelines from Human-Computer Interaction research may be helpful \cite{smith1986guidelines}.  The display should contain only immediately usable data that doesn't need to be converted, it should have consistent standards and conventions, make data fields visually distinct, tables should be used for numerical and ordinal comparisons, graphs for comparisons over space and time \cite{karjalainen2011consumer}.  The cognitive psychology of working and associative memory \cite{anderson2007can}, as well as theories of reinforcement learning \cite{sutton1998reinforcement}, suggest that immediate feedback that is contingent on behavior is the most important.  It should be ``immediate, prominent, accessible, and specific to the consumer'' \cite{roberts2004consumer}.  This is also consistent with applied behavior analysis \cite{cooper1987applied}.

\subsubsection{The Design of Everyday Things}

The first set of design principles are taken from Donald Norman's \cite{norman2002design} classic work on usable design:
\begin{itemize}
\item {\bf Use natural mappings}
\item {\bf Provide a good conceptual model}
\item {\bf Make things visible}
\item {\bf Use Affordances}
\item {\bf Use constraints}
\item {\bf Minimal and reversible errors}
\end{itemize}

Natural mappings of intended actions to functions on the device are obvious, apparent, and require no thought.  The ``the controls are just where they ought to be''.  A design with labels is not a natural mapping,``If the design depends on labels, it may be faulty...wherever labels seem necessary, consider another design'' (Norman, 1988).  Natural mappings are usually one-to-one, where each action one can take corresponds to one and only one function the device can provide.  One-to one mapping between buttons and visual cues and activities probably the best.  Natural mappings should make desires and intentions related to the device readily apparent and obviously possible or not possible.  Thus, a one-to-one mapping system results in ``natural relationships between the controls and the things controlled'' (Norman, 1988), where ``single controls have single functions''.  

There is what the designer intends (the designer's model), what the user perceives (the user's mental or conceptual model) and what is actually apparent in the device (the system image).  A good conceptual model matches the designer's model and user's model through the system image.  It ``allows you to predict the effect of your actions'' (Norman, 1988).  A good conceptual modoel is a simple causal model can be derived by merely looking at the device.  ``consistency in the presentation of operations and results and a coherent, consistent system image'' (Norman, 1988). 

Making things visible means that the important parts of the device that the person interacts with, must read, or must use to determine the system state, are visible.  A natural design yeilds a natural interpretation of a visible cue.  Can the person tell the state of the device and action alternatives by just looking at the device?  Make relevants part visible.

Affordances mean that the the user knows what to do just by looking at it.

Constraints means the user cannot do unproductive things; cosntraints limit the possible affordances.  Constraints can be physical, semantic, cultural, and logical.  The number of possible actions is physically limited by what will physically work. ``Semantic constraints use our knowledge of the situation and the world'' (Norman, 1988).  Cultural constraints are norms or conventions often used in the form of schemas (scripts; \citet{shank1977scripts}; behavior frames in Frame analysis; \citet{goffman1974frame}).  Natural mappings are logical constraints (e.g., let burner is logically connected with left knob), ``logical relationship between spatial or functional layout and the things that they affect or are affected by.'' 

Minimal and Reversible errors mean that ``if an error impsossible, someone will make it. The designer must assume that all possible errors will occur and design so as to minimize the chance of the error in the first place, or its effects once it gets made.  Errors should be easy to detect, they should have minimal consquences, and, if possible, their effects should be reversible'' (Norman, 1988).  Errors can be slips and mistakes. Slips are automatic whereas mistakes are deliberative and conscious.  There are six types of slips: 
\begin{itemize}
\item {\bf Capture errors}
\item {\bf Description errors}
\item {\bf Data-driven errors}
\item {\bf Associatiove activation errors}
\item {\bf Loss-of-activation errors}
\item {\bf Mode errors}
\end{itemize}
  
Capture error: two different action sequences have the same initial sequence and one gets confused with the other (e.g., starting to play one song and finishing another).  Imprecise description of an action leads to the wrong one being performed (e.g., throwing shirt in the toilet); different actions have similar descriptions. Data driven activities impinge on an ongoing action sequence. Associative errors are due to associations in memory.  Loss of activation errors are where you forget what you were doing; forgeting parts of actions.  Mode errors “occur when devices have different modes of operation, and the action appropriate for one mode has different meanings in other modes” (Norman, 1988; pg. 110).  Mistakes are usually choosing incorrect goals: poor decision, misclassification of a situation, failure to take all the relevant factors into account. 

\begin{enumerate}
\item Understand the causes of error and design to minimize the causes
\item Make it possible to reverse actions
\item Make it easier to discover errrors that occur and correct them
\item Use forcing functions to prevent error: interlocking; lockins; lockouts
\item Assume that every possible mishap will happen, so protect against it; make actions reversible and less costly
\item Put knowledge in the world; don't make it all be in the head
\item Use the 4 types of constraints, natural mappings, forcing functions (where necessary) 
\item Narrow the gulfs of execution and evaluation 
  \begin{itemize}
\item Execution: make options readily available
\item Evaluation: make it possible to determine system state readily easily and accurately; consistent with goals, intentions and expectations 
\end{itemize} 
\end{enumerate}

\subsubsection{Krug's Laws}

Krug's \cite{krug2009don} three laws of usability for web design are helpful:

\begin{itemize}
\item {\bf First Law}:  Don't make the user think.  
\item {\bf Second Law}: As long as actions are mindless, it doens't matter how many there are. 
\item {\bf Third Law}: Omit needless words.
\end{itemize}

According to these laws, you should be able to 'get it' without having to think about it. Things should be obviously interactive or obviously not, not in-between.  If these principles are followed, the device will “work most of the magic at a glance” \cite{krug2009don}.  This is good, because people look at it much less than we think.   Clicks that are 'on the right track' and have low uncertainty/ambiguity in the result.  Remove half the words, then remove half of that. Kill happy talk (introductions). No need for instructions; if people cant' muddle through, they won't bother reading the instructions.  No need for introductions and instructions.  Get rid of as many words as possible (omit needless words). 

The important areas for learning are cost and electricity use of different appliances, total costs, baseline consumption, the range of normal use, effective savings behaviors, and temporal patterns of use \cite{anderson2009exploring}.  To act on this learning they can do five things: 1) turn it off, 2) use it less, 3) use it more carefully, 4) improve its performance, or 5) replace it/use an alternative \cite{anderson2009exploring}.  The display can help both understanding short-term and long-term habits \cite{anderson2009exploring}.  It can help clarify the net benefits and possible actions for the electricity consumer \cite{anderson2009exploring}.

\subsection{Specific Design Elements}

Unfortunately, general principles only get us so far.  However, we do not have a blank slate with respect to the minute details of the design of the in-home display.  In this section, we discuss existing research that can be used to help with the minutia of the design.

Due to working memory limitations, people are unlikely to be able to understand more than three `bits' or `chunks' of information on the display (LeCompte, 1999; Wood and Newborough, 2003).  It is ambiguous how novices will chunk the information on the display (e.g., by single bars on a bar graph, or groups of bars) (Chase and Simon, 1973).  Thus, a cautious approach would only put three bars for a bar graph, or three numbers for a numerical table. Wood and Newborough (2003) limited each display to 3 moving digits down to kW, 1/10 kW, and 1/100 kW, but this assumes people store three chunks for each display, rather than three chunks total.

It is possible that people will be able to use the range or probability density function of use, ``A particular strength of the speedometer identified by the group was that 'you can see where you are coming from and where you are going to', unlike a digital numeric display.'' (Anderson and White, 2009) It may also be important to provide a range of uses that people typically use, or a probability density function (Wood and Newborough, 2007).

A main page or home page gives people something to hold onto \cite{krug2009don}.  The home page should make clear what device is and its purpose by answering the four big questions clearly, unambiguously, at a glance, and with little effort: 1) what is this? 2) What can I do here? 3) what do they have here?, 4) Why should I be here---and not somewhere else?  A 6--8 word tagline with a welcome blurb also may be effective.

A clear visual hierarchy should be used, such that the more important something is, the more prominent it is (e.g., size); logically related items should be visually grouped \cite{krug2009don}.  Persistent navigation on every page means that the person can tell where they are, and 'breadcrumbs' tell them how to get back to the home page.  Pages should be broken up into clearly defined areas, where one knows exactly what to expect from each area, and what interactivity is possible.  Any noise, busyness, or unnecessary background elements hsould be removed, ``Everything is visual noise until proven otherwise.''

The device should also take advantage of conventions \cite{krug2009don}, by working similar to existing media that display similar information, such as newspapers and webpages.  By doing this, the user already knows what to expect, because they have interacted with these more conventional devices quite often.

The size and legibility of text matters, movement or stationarity, and colors (a maximum of six) can be useful for encoding information. (Wood and Newborough, 2003).  Wood and Newborough (2003) used a colored digital display designed for viewing at 400-700mm distance readable up to 2m with an aspect ratio of 0.7:1.  

Users may not be able to detect small changes in kWh or frequencies of change faster than 200ms, and the scale of usage ranges from tenths of a kWh to multiple kWh per day (Wood and Newborough, 2003).   Frequencies of change between 8-12 seconds are very likely to be detected.  

Wood and Newborough (2003) used current electricity use with a reset button to measure use of a specific task or event at the top left of the screen, along with daily consumption comparied to yesterday, and weekly consumption compared to last week.  A button to help cycle through time periods may be desirable, as well as a button to help set targets that changes color if one is projected to be off-target (Anderson and White, 2009). With color feedback, a simple green or red light can indicate whether one is on or off-track for any targets.  Then the person can query the device to find out why they are off-target, and the device can offer suggestions for how to get back on.  

Wood and Newborough (2007) suggest matching the information type to the group of appliances, with low automation/high setting appliances focusing on high-frequency `event-driven' feedback, low automation low setting appliances with cumulative frequency information, and automated appliances with low frequency event driven reports.

Wood and Newborough (2003) discuss three ``types of reading'' that may take place:\begin{enumerate}
  \item Quantitative Reading: Looking for numbers
  \item Check Reading: Looking for changes and rates
  \item Setting: Observing the relationship between actions and display values
 \end{enumerate}

Different elements of the display score higher or lower on each type of reading, for example, a counter is good for quantitative reading and setting, but not qualitative reading (Carveth and Adams, 77, 78, 79; Wood and Newborough, 2003).

There is also the question of how the information is displayed.  Bar charts with three or fewer bars are likely to be highly effective (Roberts and Baker, 2003; Wood and Newborough, 2007).  

The scaling of time periods and intervals on the graphs (e.g., logarithmic vs. linear) are very important.  The spacing between bars matters and should be smaller than the width of the bars (Wood and Newborough, 2007).  

It is also important to note the reference frame in which information is displayed.  Space heating compared to lights will seem much larger, both in kWh and cost, and this may demotivate people to care about changing their behavior with respect to lights.  This could also occur on a room-by-room or person-by-person basis (Wood and Newborough, 2007).  

It might be important to use weather-normalized or ‘smoothed’ consumption values for comparison instead of raw numbers (Karjalainen, 2011).  There is a problem when electricity use spikes and dips in real-time because it can distract householders from identifying those appliances that contribute most to usage over time. There is a distinct problem with real-time monitors that show the size of instantaneous demand and then extrapolate it to give wildly differing costs within the space of a few minutes, according to what is switched on and functioning at a given moment. What is the user to believe?” (Darby, 2010).  However, weather, price and other corrections are not easily comprehended (Darby, 2010).

Prospective memory is remembering to remember, and reminders help this.  Reminders should signal that something is to be remembered, at the appropriate time, and what that thing is (Norman, 1988).  

A device with response feedback ensures that every action the user takes is reflected immediately and obviously on the device, so she can be sure her actions had an effect (whehter it was intended or not) (Norman, 1988).

\subsection{Heterogenous Agents}

Just as appliances differ in the power required to operate them, people differ in their responsiveness to feedback.  There may be multiple types of people who interact with the device, such as ``Seekers, detectives, and judges'' \cite{van2010home}.  Some types of people may already be at their limits and cannot save more, whereas others may be able to save but become become frustrated or anxious when difficulty arises (Darby, 2010).  The design of the display may be tailored to the individual or the group based on norms associated with regional differences, such as different areas in north America or Europe (Lopes et al, 2012).  High awareness participants wanted more specific feedback, such as appliance-specific information  (Yun, 2009).
  
It may be important for the interface to adapt to the knowledge of the user, for example, initially providing simple tips, but then displaying more sophisticated graphs once the user is more knowledgeable (Wood and Newborough, 2003).  For example, Paetz et al (2012) found a progression from looking at single appliance use when the participant was a beginner to looking at total household use when the participant was an expert “Now that I know all about the coffee machine and the TV, I’m more interested to know how much we have consumed over the last week or over the last month.” (T1 interview)” (Paetz et al, 2012)

\subsection{Frequency}
Black and Dylan, 1998 
Kluger and Denisi, 1996
Mayer and Clark, 2011

Do people even want in-home displays, or would they prefer feedback by email, postal mail, the internet or on a smart phone?  Many studies show people prefer in-home displays over web-based feedback, emails, mobile phones, or bill inserts (Vassileva, Odlare, Wallin, Dahlquist, 2012).  However, this varies by demographics, where low consumers may prefer email, whereas high consumers prefer web or letters (Vassileva, Odlare, Wallin, Dahlquist, 2012).  Those in apartments and houses may have different preferences for feedback information, with those in apartments preferring in-home displays but those in houses preferring email or web, possibly driven by age (e.g., younger people are more likely to live in apartments and like in-homme display technology), occupation, and income (Vassileva, Odlare, Wallin and Dahlquist, 2012).  Smart-phones are consistently seen as the least preferred way of receiving feedback.  Lower income people prefer letters and websites, moderate income prefer email, and high income people prefer a web-site (Vassileva, Odlare, Wallin and Dahlquist, 2012).  Displays are preferred by middle-income people, whereas email and websites are preferred by low or high income people (Vassileva, Wallin, and Dahlquist, 2012).  Older adults may prefer in-home displays because ``they seem easier to use and there is no Internet/computers involved'' (Vassileva, Wallin, and Dahlquist, 2012).  This also breaks down by house vs. apartment, where displays are chosen more by those living in apartments (usually younger more tech-savvy people) and email and websites preferred more for houses.  Looking up web-based feedback may require more effort than using a display, and use of web-based information is still quite low in the U.S and Sweden (2-4\%) (Darby, 2010). 

\cite{wood2007energy} suggest sectioning appliances by the degree of automation and possible settings they can take, with cookers having low automation and high number of settings being the best target for energy use information.  Different groups of appliances could have their information aggregated together in the display.  This is also called ``degree of user involvement'' of the appliance \cite{bladh2008towards}.  These include grouping appliances, background and foreground distinctions \cite{wood2003dynamic}.

There also seemed to be individual differences in `money-saving attitude' that appeared to be connected to desire for feedback.  People who perceive their consumption to be minimal do not want feedback.







